Merge tag 'drm-next-2019-09-18' of git://anongit.freedesktop.org/drm/drm

Pull drm updates from Dave Airlie:
 "This is the main pull request for 5.4-rc1 merge window. I don't think
  there is anything outstanding so next week should just be fixes, but
  we'll see if I missed anything. I landed some fixes earlier in the
  week but got delayed writing summary and sending it out, due to a mix
  of sick kid and jetlag!

  There are some fixes pending, but I'd rather get the main merge out of
  the way instead of delaying it longer.

  It's also pretty large in commit count and new amd header file size.
  The largest thing is four new amdgpu products (navi12/14, arcturus and
  renoir APU support).

  Otherwise it's pretty much lots of work across the board, i915 has
  started landing tigerlake support, lots of icelake fixes and lots of
  locking reworking for future gpu support, lots of header file rework
  (drmP.h is nearly gone), some old legacy hacks (DRM_WAIT_ON) have been
  put into the places they are needed.

  uapi:
   - content protection type property for HDCP

  core:
   - rework include dependencies
   - lots of drmP.h removals
   - link rate calculation robustness fix
   - make fb helper map only when required
   - add connector->DDC adapter link
   - DRM_WAIT_ON removed
   - drop DRM_AUTH usage from drivers

  dma-buf:
   - reservation object fence helper

  dma-fence:
   - shrink dma_fence struct
   - merge signal functions
   - store timestamps in dma_fence
   - selftests

  ttm:
   - embed drm_get_object struct into ttm_buffer_object
   - release_notify callback

  bridges:
   - sii902x - audio graph card support
   - tc358767 - aux data handling rework
   - ti-snd64dsi86 - debugfs support, DSI mode flags support

  panels:
   - Support for GiantPlus GPM940B0, Sharp LQ070Y3DG3B, Ortustech
     COM37H3M, Novatek NT39016, Sharp LS020B1DD01D, Raydium RM67191, Boe
     Himax8279d, Sharp LD-D5116Z01B
   - TI nspire, NEC NL8048HL11, LG Philips LB035Q02, Sharp LS037V7DW01,
     Sony ACX565AKM, Toppoly TD028TTEC1 Toppoly TD043MTEA1

  i915:
   - Initial tigerlake platform support
   - Locking simplification work, general all over refactoring.
   - Selftests
   - HDCP debug info improvements
   - DSI properties
   - Icelake display PLL fixes, colorspace fixes, bandwidth fixes, DSI
     suspend/resume
   - GuC fixes
   - Perf fixes
   - ElkhartLake enablement
   - DP MST fixes
   - GVT - command parser enhancements

  amdgpu:
   - add wipe memory on release flag for buffer creation
   - Navi12/14 support (may be marked experimental)
   - Arcturus support
   - Renoir APU support
   - mclk DPM for Navi
   - DC display fixes
   - Raven scatter/gather support
   - RAS support for GFX
   - Navi12 + Arcturus power features
   - GPU reset for Picasso
   - smu11 i2c controller support

  amdkfd:
   - navi12/14 support
   - Arcturus support

  radeon:
   - kexec fix

  nouveau:
   - improved display color management
   - detect lack of GPU power cables

  vmwgfx:
   - evicition priority support
   - remove unused security feature

  msm:
   - msm8998 display support
   - better async commit support for cursor updates

  etnaviv:
   - per-process address space support
   - performance counter fixes
   - softpin support

  mcde:
   - DCS transfers fix

  exynos:
   - drmP.h cleanup

  lima:
   - reduce logging

  kirin:
   - misc clenaups

  komeda:
   - dual-link support
   - DT memory regions

  hisilicon:
   - misc fixes

  imx:
   - IPUv3 image converter fixes
   - 32-bit RGB V4L2 pixel format support

  ingenic:
   - more support for panel related cases

  mgag200:
   - cursor support fix

  panfrost:
   - export GPU features register to userspace
   - gpu heap allocations
   - per-fd address space support

  pl111:
   - CLD pads wiring support removed from DT

  rockchip:
   - rework to use DRM PSR helpers
   - fix bug in VOP_WIN_GET macro
   - DSI DT binding rework

  sun4i:
   - improve support for color encoding and range
   - DDC enabled GPIO

  tinydrm:
   - rework SPI support
   - improve MIPI-DBI support
   - moved to drm/tiny

  vkms:
   - rework CRC tracking

  dw-hdmi:
   - get_eld and i2s improvements

  gm12u320:
   - misc fixes

  meson:
   - global code cleanup
   - vpu feature detect

  omap:
   - alpha/pixel blend mode properties

  rcar-du:
   - misc fixes"

* tag 'drm-next-2019-09-18' of git://anongit.freedesktop.org/drm/drm: (2112 commits)
  drm/nouveau/bar/gm20b: Avoid BAR1 teardown during init
  drm/nouveau: Fix ordering between TTM and GEM release
  drm/nouveau/prime: Extend DMA reservation object lock
  drm/nouveau: Fix fallout from reservation object rework
  drm/nouveau/kms/nv50-: Don't create MSTMs for eDP connectors
  drm/i915: Use NOEVICT for first pass on attemping to pin a GGTT mmap
  drm/i915: to make vgpu ppgtt notificaiton as atomic operation
  drm/i915: Flush the existing fence before GGTT read/write
  drm/i915: Hold irq-off for the entire fake lock period
  drm/i915/gvt: update RING_START reg of vGPU when the context is submitted to i915
  drm/i915/gvt: update vgpu workload head pointer correctly
  drm/mcde: Fix DSI transfers
  drm/msm: Use the correct dma_sync calls harder
  drm/msm: remove unlikely() from WARN_ON() conditions
  drm/msm/dsi: Fix return value check for clk_get_parent
  drm/msm: add atomic traces
  drm/msm/dpu: async commit support
  drm/msm: async commit support
  drm/msm: split power control from prepare/complete_commit
  drm/msm: add kms->flush_commit()
  ...

20 files changed, 2336 insertions, 2265 deletions

diff --git a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h
index 826913c70766..a8cf82d46109 100644
--- a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h
+++ b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h
@@ -274,154 +274,227 @@ static const uint32_t cwsr_trap_gfx8_hex[] = {
 
 
 static const uint32_t cwsr_trap_gfx9_hex[] = {
-	0xbf820001, 0xbf82015e,
+	0xbf820001, 0xbf820248,
 	0xb8f8f802, 0x89788678,
-	0xb8fbf803, 0x866eff7b,
-	0x00000400, 0xbf85003b,
-	0x866eff7b, 0x00000800,
-	0xbf850003, 0x866eff7b,
-	0x00000100, 0xbf84000c,
+	0xb8eef801, 0x866eff6e,
+	0x00000800, 0xbf840003,
 	0x866eff78, 0x00002000,
-	0xbf840005, 0xbf8e0010,
-	0xb8eef803, 0x866eff6e,
-	0x00000400, 0xbf84fffb,
-	0x8778ff78, 0x00002000,
-	0x80ec886c, 0x82ed806d,
-	0xb8eef807, 0x866fff6e,
-	0x001f8000, 0x8e6f8b6f,
-	0x8977ff77, 0xfc000000,
-	0x87776f77, 0x896eff6e,
-	0x001f8000, 0xb96ef807,
-	0xb8faf812, 0xb8fbf813,
-	0x8efa887a, 0xc0071bbd,
-	0x00000000, 0xbf8cc07f,
-	0xc0071ebd, 0x00000008,
-	0xbf8cc07f, 0x86ee6e6e,
-	0xbf840001, 0xbe801d6e,
-	0xb8fbf803, 0x867bff7b,
-	0x000001ff, 0xbf850002,
-	0x806c846c, 0x826d806d,
+	0xbf840016, 0xb8fbf803,
+	0x866eff7b, 0x00000400,
+	0xbf85003b, 0x866eff7b,
+	0x00000800, 0xbf850003,
+	0x866eff7b, 0x00000100,
+	0xbf84000c, 0x866eff78,
+	0x00002000, 0xbf840005,
+	0xbf8e0010, 0xb8eef803,
+	0x866eff6e, 0x00000400,
+	0xbf84fffb, 0x8778ff78,
+	0x00002000, 0x80ec886c,
+	0x82ed806d, 0xb8eef807,
+	0x866fff6e, 0x001f8000,
+	0x8e6f8b6f, 0x8977ff77,
+	0xfc000000, 0x87776f77,
+	0x896eff6e, 0x001f8000,
+	0xb96ef807, 0xb8faf812,
+	0xb8fbf813, 0x8efa887a,
+	0xc0071bbd, 0x00000000,
+	0xbf8cc07f, 0xc0071ebd,
+	0x00000008, 0xbf8cc07f,
+	0x86ee6e6e, 0xbf840001,
+	0xbe801d6e, 0xb8fbf803,
+	0x867bff7b, 0x000001ff,
+	0xbf850002, 0x806c846c,
+	0x826d806d, 0x866dff6d,
+	0x0000ffff, 0x8f6e8b77,
+	0x866eff6e, 0x001f8000,
+	0xb96ef807, 0x86fe7e7e,
+	0x86ea6a6a, 0x8f6e8378,
+	0xb96ee0c2, 0xbf800002,
+	0xb9780002, 0xbe801f6c,
 	0x866dff6d, 0x0000ffff,
-	0x8f6e8b77, 0x866eff6e,
-	0x001f8000, 0xb96ef807,
-	0x86fe7e7e, 0x86ea6a6a,
-	0x8f6e8378, 0xb96ee0c2,
-	0xbf800002, 0xb9780002,
-	0xbe801f6c, 0x866dff6d,
-	0x0000ffff, 0xbefa0080,
-	0xb97a0283, 0xb8fa2407,
-	0x8e7a9b7a, 0x876d7a6d,
-	0xb8fa03c7, 0x8e7a9a7a,
-	0x876d7a6d, 0xb8faf807,
-	0x867aff7a, 0x00007fff,
-	0xb97af807, 0xbeee007e,
-	0xbeef007f, 0xbefe0180,
-	0xbf900004, 0x877a8478,
-	0xb97af802, 0xbf8e0002,
-	0xbf88fffe, 0xb8fa2a05,
-	0x807a817a, 0x8e7a8a7a,
-	0xb8fb1605, 0x807b817b,
-	0x8e7b867b, 0x807a7b7a,
-	0x807a7e7a, 0x827b807f,
-	0x867bff7b, 0x0000ffff,
-	0xc04b1c3d, 0x00000050,
-	0xbf8cc07f, 0xc04b1d3d,
-	0x00000060, 0xbf8cc07f,
-	0xc0431e7d, 0x00000074,
-	0xbf8cc07f, 0xbef4007e,
-	0x8675ff7f, 0x0000ffff,
-	0x8775ff75, 0x00040000,
-	0xbef60080, 0xbef700ff,
-	0x00807fac, 0x867aff7f,
-	0x08000000, 0x8f7a837a,
-	0x87777a77, 0x867aff7f,
-	0x70000000, 0x8f7a817a,
-	0x87777a77, 0xbef1007c,
-	0xbef00080, 0xb8f02a05,
-	0x80708170, 0x8e708a70,
-	0xb8fa1605, 0x807a817a,
-	0x8e7a867a, 0x80707a70,
-	0xbef60084, 0xbef600ff,
-	0x01000000, 0xbefe007c,
-	0xbefc0070, 0xc0611c7a,
-	0x0000007c, 0xbf8cc07f,
-	0x80708470, 0xbefc007e,
+	0xbefa0080, 0xb97a0283,
+	0xb8fa2407, 0x8e7a9b7a,
+	0x876d7a6d, 0xb8fa03c7,
+	0x8e7a9a7a, 0x876d7a6d,
+	0xb8faf807, 0x867aff7a,
+	0x00007fff, 0xb97af807,
+	0xbeee007e, 0xbeef007f,
+	0xbefe0180, 0xbf900004,
+	0x877a8478, 0xb97af802,
+	0xbf8e0002, 0xbf88fffe,
+	0xb8fa2a05, 0x807a817a,
+	0x8e7a8a7a, 0xb8fb1605,
+	0x807b817b, 0x8e7b867b,
+	0x807a7b7a, 0x807a7e7a,
+	0x827b807f, 0x867bff7b,
+	0x0000ffff, 0xc04b1c3d,
+	0x00000050, 0xbf8cc07f,
+	0xc04b1d3d, 0x00000060,
+	0xbf8cc07f, 0xc0431e7d,
+	0x00000074, 0xbf8cc07f,
+	0xbef4007e, 0x8675ff7f,
+	0x0000ffff, 0x8775ff75,
+	0x00040000, 0xbef60080,
+	0xbef700ff, 0x00807fac,
+	0x867aff7f, 0x08000000,
+	0x8f7a837a, 0x87777a77,
+	0x867aff7f, 0x70000000,
+	0x8f7a817a, 0x87777a77,
+	0xbef1007c, 0xbef00080,
+	0xb8f02a05, 0x80708170,
+	0x8e708a70, 0xb8fa1605,
+	0x807a817a, 0x8e7a867a,
+	0x80707a70, 0xbef60084,
+	0xbef600ff, 0x01000000,
 	0xbefe007c, 0xbefc0070,
-	0xc0611b3a, 0x0000007c,
+	0xc0611c7a, 0x0000007c,
 	0xbf8cc07f, 0x80708470,
 	0xbefc007e, 0xbefe007c,
-	0xbefc0070, 0xc0611b7a,
+	0xbefc0070, 0xc0611b3a,
 	0x0000007c, 0xbf8cc07f,
 	0x80708470, 0xbefc007e,
 	0xbefe007c, 0xbefc0070,
-	0xc0611bba, 0x0000007c,
+	0xc0611b7a, 0x0000007c,
 	0xbf8cc07f, 0x80708470,
 	0xbefc007e, 0xbefe007c,
-	0xbefc0070, 0xc0611bfa,
+	0xbefc0070, 0xc0611bba,
 	0x0000007c, 0xbf8cc07f,
 	0x80708470, 0xbefc007e,
 	0xbefe007c, 0xbefc0070,
-	0xc0611e3a, 0x0000007c,
-	0xbf8cc07f, 0x80708470,
-	0xbefc007e, 0xb8fbf803,
-	0xbefe007c, 0xbefc0070,
-	0xc0611efa, 0x0000007c,
+	0xc0611bfa, 0x0000007c,
 	0xbf8cc07f, 0x80708470,
 	0xbefc007e, 0xbefe007c,
-	0xbefc0070, 0xc0611a3a,
+	0xbefc0070, 0xc0611e3a,
+	0x0000007c, 0xbf8cc07f,
+	0x80708470, 0xbefc007e,
+	0xb8fbf803, 0xbefe007c,
+	0xbefc0070, 0xc0611efa,
 	0x0000007c, 0xbf8cc07f,
 	0x80708470, 0xbefc007e,
 	0xbefe007c, 0xbefc0070,
-	0xc0611a7a, 0x0000007c,
-	0xbf8cc07f, 0x80708470,
-	0xbefc007e, 0xb8f1f801,
-	0xbefe007c, 0xbefc0070,
-	0xc0611c7a, 0x0000007c,
+	0xc0611a3a, 0x0000007c,
 	0xbf8cc07f, 0x80708470,
-	0xbefc007e, 0x867aff7f,
-	0x04000000, 0xbeef0080,
-	0x876f6f7a, 0xb8f02a05,
+	0xbefc007e, 0xbefe007c,
+	0xbefc0070, 0xc0611a7a,
+	0x0000007c, 0xbf8cc07f,
+	0x80708470, 0xbefc007e,
+	0xb8f1f801, 0xbefe007c,
+	0xbefc0070, 0xc0611c7a,
+	0x0000007c, 0xbf8cc07f,
+	0x80708470, 0xbefc007e,
+	0x867aff7f, 0x04000000,
+	0xbeef0080, 0x876f6f7a,
+	0xb8f02a05, 0x80708170,
+	0x8e708a70, 0xb8fb1605,
+	0x807b817b, 0x8e7b847b,
+	0x8e76827b, 0xbef600ff,
+	0x01000000, 0xbef20174,
+	0x80747074, 0x82758075,
+	0xbefc0080, 0xbf800000,
+	0xbe802b00, 0xbe822b02,
+	0xbe842b04, 0xbe862b06,
+	0xbe882b08, 0xbe8a2b0a,
+	0xbe8c2b0c, 0xbe8e2b0e,
+	0xc06b003a, 0x00000000,
+	0xbf8cc07f, 0xc06b013a,
+	0x00000010, 0xbf8cc07f,
+	0xc06b023a, 0x00000020,
+	0xbf8cc07f, 0xc06b033a,
+	0x00000030, 0xbf8cc07f,
+	0x8074c074, 0x82758075,
+	0x807c907c, 0xbf0a7b7c,
+	0xbf85ffe7, 0xbef40172,
+	0xbef00080, 0xbefe00c1,
+	0xbeff00c1, 0xbee80080,
+	0xbee90080, 0xbef600ff,
+	0x01000000, 0x867aff78,
+	0x00400000, 0xbf850003,
+	0xb8faf803, 0x897a7aff,
+	0x10000000, 0xbf85004d,
+	0xbe840080, 0xd2890000,
+	0x00000900, 0x80048104,
+	0xd2890001, 0x00000900,
+	0x80048104, 0xd2890002,
+	0x00000900, 0x80048104,
+	0xd2890003, 0x00000900,
+	0x80048104, 0xc069003a,
+	0x00000070, 0xbf8cc07f,
+	0x80709070, 0xbf06c004,
+	0xbf84ffee, 0xbe840080,
+	0xd2890000, 0x00000901,
+	0x80048104, 0xd2890001,
+	0x00000901, 0x80048104,
+	0xd2890002, 0x00000901,
+	0x80048104, 0xd2890003,
+	0x00000901, 0x80048104,
+	0xc069003a, 0x00000070,
+	0xbf8cc07f, 0x80709070,
+	0xbf06c004, 0xbf84ffee,
+	0xbe840080, 0xd2890000,
+	0x00000902, 0x80048104,
+	0xd2890001, 0x00000902,
+	0x80048104, 0xd2890002,
+	0x00000902, 0x80048104,
+	0xd2890003, 0x00000902,
+	0x80048104, 0xc069003a,
+	0x00000070, 0xbf8cc07f,
+	0x80709070, 0xbf06c004,
+	0xbf84ffee, 0xbe840080,
+	0xd2890000, 0x00000903,
+	0x80048104, 0xd2890001,
+	0x00000903, 0x80048104,
+	0xd2890002, 0x00000903,
+	0x80048104, 0xd2890003,
+	0x00000903, 0x80048104,
+	0xc069003a, 0x00000070,
+	0xbf8cc07f, 0x80709070,
+	0xbf06c004, 0xbf84ffee,
+	0xbf820008, 0xe0724000,
+	0x701d0000, 0xe0724100,
+	0x701d0100, 0xe0724200,
+	0x701d0200, 0xe0724300,
+	0x701d0300, 0xbefe00c1,
+	0xbeff00c1, 0xb8fb4306,
+	0x867bc17b, 0xbf840063,
+	0xbf8a0000, 0x867aff6f,
+	0x04000000, 0xbf84005f,
+	0x8e7b867b, 0x8e7b827b,
+	0xbef6007b, 0xb8f02a05,
 	0x80708170, 0x8e708a70,
-	0xb8fb1605, 0x807b817b,
-	0x8e7b847b, 0x8e76827b,
-	0xbef600ff, 0x01000000,
-	0xbef20174, 0x80747074,
-	0x82758075, 0xbefc0080,
-	0xbf800000, 0xbe802b00,
-	0xbe822b02, 0xbe842b04,
-	0xbe862b06, 0xbe882b08,
-	0xbe8a2b0a, 0xbe8c2b0c,
-	0xbe8e2b0e, 0xc06b003a,
-	0x00000000, 0xbf8cc07f,
-	0xc06b013a, 0x00000010,
-	0xbf8cc07f, 0xc06b023a,
-	0x00000020, 0xbf8cc07f,
-	0xc06b033a, 0x00000030,
-	0xbf8cc07f, 0x8074c074,
-	0x82758075, 0x807c907c,
-	0xbf0a7b7c, 0xbf85ffe7,
-	0xbef40172, 0xbef00080,
-	0xbefe00c1, 0xbeff00c1,
-	0xbee80080, 0xbee90080,
+	0xb8fa1605, 0x807a817a,
+	0x8e7a867a, 0x80707a70,
+	0x8070ff70, 0x00000080,
 	0xbef600ff, 0x01000000,
-	0xe0724000, 0x701d0000,
-	0xe0724100, 0x701d0100,
-	0xe0724200, 0x701d0200,
-	0xe0724300, 0x701d0300,
-	0xbefe00c1, 0xbeff00c1,
-	0xb8fb4306, 0x867bc17b,
-	0xbf84002c, 0xbf8a0000,
-	0x867aff6f, 0x04000000,
-	0xbf840028, 0x8e7b867b,
-	0x8e7b827b, 0xbef6007b,
-	0xb8f02a05, 0x80708170,
-	0x8e708a70, 0xb8fa1605,
-	0x807a817a, 0x8e7a867a,
-	0x80707a70, 0x8070ff70,
-	0x00000080, 0xbef600ff,
-	0x01000000, 0xbefc0080,
-	0xd28c0002, 0x000100c1,
-	0xd28d0003, 0x000204c1,
+	0xbefc0080, 0xd28c0002,
+	0x000100c1, 0xd28d0003,
+	0x000204c1, 0x867aff78,
+	0x00400000, 0xbf850003,
+	0xb8faf803, 0x897a7aff,
+	0x10000000, 0xbf850030,
+	0x24040682, 0xd86e4000,
+	0x00000002, 0xbf8cc07f,
+	0xbe840080, 0xd2890000,
+	0x00000900, 0x80048104,
+	0xd2890001, 0x00000900,
+	0x80048104, 0xd2890002,
+	0x00000900, 0x80048104,
+	0xd2890003, 0x00000900,
+	0x80048104, 0xc069003a,
+	0x00000070, 0xbf8cc07f,
+	0x80709070, 0xbf06c004,
+	0xbf84ffee, 0xbe840080,
+	0xd2890000, 0x00000901,
+	0x80048104, 0xd2890001,
+	0x00000901, 0x80048104,
+	0xd2890002, 0x00000901,
+	0x80048104, 0xd2890003,
+	0x00000901, 0x80048104,
+	0xc069003a, 0x00000070,
+	0xbf8cc07f, 0x80709070,
+	0xbf06c004, 0xbf84ffee,
+	0x680404ff, 0x00000200,
+	0xd0c9006a, 0x0000f702,
+	0xbf87ffd2, 0xbf820015,
 	0xd1060002, 0x00011103,
 	0x7e0602ff, 0x00000200,
 	0xbefc00ff, 0x00010000,
@@ -438,9 +511,53 @@ static const uint32_t cwsr_trap_gfx9_hex[] = {
 	0x807b817b, 0x8e7b827b,
 	0x8e76887b, 0xbef600ff,
 	0x01000000, 0xbefc0084,
-	0xbf0a7b7c, 0xbf840015,
+	0xbf0a7b7c, 0xbf84006d,
 	0xbf11017c, 0x807bff7b,
-	0x00001000, 0x7e000300,
+	0x00001000, 0x867aff78,
+	0x00400000, 0xbf850003,
+	0xb8faf803, 0x897a7aff,
+	0x10000000, 0xbf850051,
+	0xbe840080, 0xd2890000,
+	0x00000900, 0x80048104,
+	0xd2890001, 0x00000900,
+	0x80048104, 0xd2890002,
+	0x00000900, 0x80048104,
+	0xd2890003, 0x00000900,
+	0x80048104, 0xc069003a,
+	0x00000070, 0xbf8cc07f,
+	0x80709070, 0xbf06c004,
+	0xbf84ffee, 0xbe840080,
+	0xd2890000, 0x00000901,
+	0x80048104, 0xd2890001,
+	0x00000901, 0x80048104,
+	0xd2890002, 0x00000901,
+	0x80048104, 0xd2890003,
+	0x00000901, 0x80048104,
+	0xc069003a, 0x00000070,
+	0xbf8cc07f, 0x80709070,
+	0xbf06c004, 0xbf84ffee,
+	0xbe840080, 0xd2890000,
+	0x00000902, 0x80048104,
+	0xd2890001, 0x00000902,
+	0x80048104, 0xd2890002,
+	0x00000902, 0x80048104,
+	0xd2890003, 0x00000902,
+	0x80048104, 0xc069003a,
+	0x00000070, 0xbf8cc07f,
+	0x80709070, 0xbf06c004,
+	0xbf84ffee, 0xbe840080,
+	0xd2890000, 0x00000903,
+	0x80048104, 0xd2890001,
+	0x00000903, 0x80048104,
+	0xd2890002, 0x00000903,
+	0x80048104, 0xd2890003,
+	0x00000903, 0x80048104,
+	0xc069003a, 0x00000070,
+	0xbf8cc07f, 0x80709070,
+	0xbf06c004, 0xbf84ffee,
+	0x807c847c, 0xbf0a7b7c,
+	0xbf85ffb1, 0xbf9c0000,
+	0xbf820012, 0x7e000300,
 	0x7e020301, 0x7e040302,
 	0x7e060303, 0xe0724000,
 	0x701d0000, 0xe0724100,
@@ -563,24 +680,47 @@ static const uint32_t cwsr_trap_gfx9_hex[] = {
 };
 
 static const uint32_t cwsr_trap_gfx10_hex[] = {
-	0xbf820001, 0xbf82012e,
-	0xb0804004, 0xb970f802,
-	0x8a708670, 0xb971f803,
-	0x8771ff71, 0x00000400,
-	0xbf850008, 0xb971f803,
-	0x8771ff71, 0x000001ff,
-	0xbf850001, 0x806c846c,
+	0xbf820001, 0xbf8201c1,
+	0xb0804004, 0xb978f802,
+	0x8a788678, 0xb971f803,
+	0x876eff71, 0x00000400,
+	0xbf850033, 0x876eff71,
+	0x00000100, 0xbf840002,
+	0x8878ff78, 0x00002000,
+	0x8a77ff77, 0xff000000,
+	0xb96ef807, 0x876fff6e,
+	0x02000000, 0x8f6f866f,
+	0x88776f77, 0x876fff6e,
+	0x003f8000, 0x8f6f896f,
+	0x88776f77, 0x8a6eff6e,
+	0x023f8000, 0xb9eef807,
+	0xb970f812, 0xb971f813,
+	0x8ff08870, 0xf4051bb8,
+	0xfa000000, 0xbf8cc07f,
+	0xf4051c38, 0xfa000008,
+	0xbf8cc07f, 0x87ee6e6e,
+	0xbf840001, 0xbe80206e,
+	0xb971f803, 0x8771ff71,
+	0x000001ff, 0xbf850002,
+	0x806c846c, 0x826d806d,
+	0x876dff6d, 0x0000ffff,
+	0x906e8977, 0x876fff6e,
+	0x003f8000, 0x906e8677,
+	0x876eff6e, 0x02000000,
+	0x886e6f6e, 0xb9eef807,
+	0x87fe7e7e, 0x87ea6a6a,
+	0xb9f8f802, 0xbe80226c,
+	0xb971f803, 0x8771ff71,
+	0x00000100, 0xbf840006,
+	0xbef60380, 0xb9f60203,
 	0x876dff6d, 0x0000ffff,
-	0xbe80226c, 0xb971f803,
-	0x8771ff71, 0x00000100,
-	0xbf840006, 0xbef60380,
-	0xb9f60203, 0x876dff6d,
-	0x0000ffff, 0x80ec886c,
-	0x82ed806d, 0xbef60380,
-	0xb9f60283, 0xb973f816,
-	0xb9762c07, 0x8f769c76,
-	0x886d766d, 0xb97603c7,
-	0x8f769b76, 0x886d766d,
+	0x80ec886c, 0x82ed806d,
+	0xbef60380, 0xb9f60283,
+	0xb972f816, 0xb9762c07,
+	0x8f769a76, 0x886d766d,
+	0xb97603c7, 0x8f769976,
+	0x886d766d, 0xb9760647,
+	0x8f769876, 0x886d766d,
 	0xb976f807, 0x8776ff76,
 	0x00007fff, 0xb9f6f807,
 	0xbeee037e, 0xbeef037f,
@@ -589,274 +729,833 @@ static const uint32_t cwsr_trap_gfx10_hex[] = {
 	0xbef4037e, 0x8775ff7f,
 	0x0000ffff, 0x8875ff75,
 	0x00040000, 0xbef60380,
-	0xbef703ff, 0x00807fac,
+	0xbef703ff, 0x10807fac,
 	0x8776ff7f, 0x08000000,
 	0x90768376, 0x88777677,
 	0x8776ff7f, 0x70000000,
 	0x90768176, 0x88777677,
 	0xbefb037c, 0xbefa0380,
-	0xb97202dc, 0x8872727f,
-	0xbefe03c1, 0x877c8172,
-	0xbf06817c, 0xbf850002,
-	0xbeff0380, 0xbf820001,
-	0xbeff03c1, 0xb9712a05,
-	0x80718171, 0x8f718271,
-	0x877c8172, 0xbf06817c,
-	0xbf85000d, 0x8f768771,
+	0xb97302dc, 0x8f739973,
+	0x8873737f, 0xb97a2a05,
+	0x807a817a, 0x907c9973,
+	0x877c817c, 0xbf06817c,
+	0xbf850002, 0x8f7a897a,
+	0xbf820001, 0x8f7a8a7a,
+	0xb9761e06, 0x8f768a76,
+	0x807a767a, 0x807aff7a,
+	0x00000200, 0xbef603ff,
+	0x01000000, 0xbefe037c,
+	0xbefc037a, 0xf4611efa,
+	0xf8000000, 0x807a847a,
+	0xbefc037e, 0xbefe037c,
+	0xbefc037a, 0xf4611b3a,
+	0xf8000000, 0x807a847a,
+	0xbefc037e, 0xbefe037c,
+	0xbefc037a, 0xf4611b7a,
+	0xf8000000, 0x807a847a,
+	0xbefc037e, 0xbefe037c,
+	0xbefc037a, 0xf4611bba,
+	0xf8000000, 0x807a847a,
+	0xbefc037e, 0xbefe037c,
+	0xbefc037a, 0xf4611bfa,
+	0xf8000000, 0x807a847a,
+	0xbefc037e, 0xbefe037c,
+	0xbefc037a, 0xf4611e3a,
+	0xf8000000, 0x807a847a,
+	0xbefc037e, 0xb971f803,
+	0xbefe037c, 0xbefc037a,
+	0xf4611c7a, 0xf8000000,
+	0x807a847a, 0xbefc037e,
+	0xbefe037c, 0xbefc037a,
+	0xf4611cba, 0xf8000000,
+	0x807a847a, 0xbefc037e,
+	0xb97bf801, 0xbefe037c,
+	0xbefc037a, 0xf4611efa,
+	0xf8000000, 0x807a847a,
+	0xbefc037e, 0xb97bf814,
+	0xbefe037c, 0xbefc037a,
+	0xf4611efa, 0xf8000000,
+	0x807a847a, 0xbefc037e,
+	0xb97bf815, 0xbefe037c,
+	0xbefc037a, 0xf4611efa,
+	0xf8000000, 0x807a847a,
+	0xbefc037e, 0x8776ff7f,
+	0x04000000, 0xbeef0380,
+	0x886f6f76, 0xb97a2a05,
+	0x807a817a, 0x907c9973,
+	0x877c817c, 0xbf06817c,
+	0xbf850002, 0x8f7a897a,
+	0xbf820001, 0x8f7a8a7a,
+	0xb9761e06, 0x8f768a76,
+	0x807a767a, 0xbef603ff,
+	0x01000000, 0xbef20374,
+	0x80747a74, 0x82758075,
+	0xbefc0380, 0xbf800000,
+	0xbe802f00, 0xbe822f02,
+	0xbe842f04, 0xbe862f06,
+	0xbe882f08, 0xbe8a2f0a,
+	0xbe8c2f0c, 0xbe8e2f0e,
+	0xf469003a, 0xfa000000,
+	0xf469013a, 0xfa000010,
+	0xf469023a, 0xfa000020,
+	0xf469033a, 0xfa000030,
+	0x8074c074, 0x82758075,
+	0x807c907c, 0xbf0aff7c,
+	0x00000060, 0xbf85ffea,
+	0xbe802f00, 0xbe822f02,
+	0xbe842f04, 0xbe862f06,
+	0xbe882f08, 0xbe8a2f0a,
+	0xf469003a, 0xfa000000,
+	0xf469013a, 0xfa000010,
+	0xf469023a, 0xfa000020,
+	0x8074b074, 0x82758075,
+	0xbef40372, 0xbefa0380,
+	0xbefe03c1, 0x907c9973,
+	0x877c817c, 0xbf06817c,
+	0xbf850002, 0xbeff0380,
+	0xbf820002, 0xbeff03c1,
+	0xbf82000b, 0xbef603ff,
+	0x01000000, 0xe0704000,
+	0x7a5d0000, 0xe0704080,
+	0x7a5d0100, 0xe0704100,
+	0x7a5d0200, 0xe0704180,
+	0x7a5d0300, 0xbf82000a,
 	0xbef603ff, 0x01000000,
-	0xbefc0380, 0x7e008700,
-	0xe0704000, 0x7a5d0000,
-	0x807c817c, 0x807aff7a,
-	0x00000080, 0xbf0a717c,
-	0xbf85fff8, 0xbf82001b,
-	0x8f768871, 0xbef603ff,
-	0x01000000, 0xbefc0380,
-	0x7e008700, 0xe0704000,
-	0x7a5d0000, 0x807c817c,
-	0x807aff7a, 0x00000100,
-	0xbf0a717c, 0xbf85fff8,
-	0xb9711e06, 0x8771c171,
-	0xbf84000c, 0x8f718371,
-	0x80717c71, 0xbefe03c1,
-	0xbeff0380, 0x7e008700,
 	0xe0704000, 0x7a5d0000,
-	0x807c817c, 0x807aff7a,
-	0x00000080, 0xbf0a717c,
-	0xbf85fff8, 0xbf8a0000,
-	0x8776ff72, 0x04000000,
-	0xbf84002b, 0xbefe03c1,
-	0x877c8172, 0xbf06817c,
+	0xe0704100, 0x7a5d0100,
+	0xe0704200, 0x7a5d0200,
+	0xe0704300, 0x7a5d0300,
+	0xbefe03c1, 0x907c9973,
+	0x877c817c, 0xbf06817c,
 	0xbf850002, 0xbeff0380,
 	0xbf820001, 0xbeff03c1,
 	0xb9714306, 0x8771c171,
-	0xbf840021, 0x8f718671,
+	0xbf840046, 0xbf8a0000,
+	0x8776ff6f, 0x04000000,
+	0xbf840042, 0x8f718671,
 	0x8f718271, 0xbef60371,
+	0xb97a2a05, 0x807a817a,
+	0x907c9973, 0x877c817c,
+	0xbf06817c, 0xbf850002,
+	0x8f7a897a, 0xbf820001,
+	0x8f7a8a7a, 0xb9761e06,
+	0x8f768a76, 0x807a767a,
+	0x807aff7a, 0x00000200,
+	0x807aff7a, 0x00000080,
 	0xbef603ff, 0x01000000,
 	0xd7650000, 0x000100c1,
 	0xd7660000, 0x000200c1,
-	0x16000084, 0x877c8172,
-	0xbf06817c, 0xbefc0380,
-	0xbf85000a, 0x807cff7c,
-	0x00000080, 0x807aff7a,
-	0x00000080, 0xd5250000,
-	0x0001ff00, 0x00000080,
-	0xbf0a717c, 0xbf85fff7,
-	0xbf820009, 0x807cff7c,
-	0x00000100, 0x807aff7a,
-	0x00000100, 0xd5250000,
-	0x0001ff00, 0x00000100,
-	0xbf0a717c, 0xbf85fff7,
-	0x877c8172, 0xbf06817c,
-	0xbf850003, 0x8f7687ff,
-	0x0000006a, 0xbf820002,
-	0x8f7688ff, 0x0000006a,
+	0x16000084, 0x907c9973,
+	0x877c817c, 0xbf06817c,
+	0xbefc0380, 0xbf850012,
+	0xbe8303ff, 0x00000080,
+	0xbf800000, 0xbf800000,
+	0xbf800000, 0xd8d80000,
+	0x01000000, 0xbf8c0000,
+	0xe0704000, 0x7a5d0100,
+	0x807c037c, 0x807a037a,
+	0xd5250000, 0x0001ff00,
+	0x00000080, 0xbf0a717c,
+	0xbf85fff4, 0xbf820011,
+	0xbe8303ff, 0x00000100,
+	0xbf800000, 0xbf800000,
+	0xbf800000, 0xd8d80000,
+	0x01000000, 0xbf8c0000,
+	0xe0704000, 0x7a5d0100,
+	0x807c037c, 0x807a037a,
+	0xd5250000, 0x0001ff00,
+	0x00000100, 0xbf0a717c,
+	0xbf85fff4, 0xbefe03c1,
+	0x907c9973, 0x877c817c,
+	0xbf06817c, 0xbf850004,
+	0xbefa03ff, 0x00000200,
+	0xbeff0380, 0xbf820003,
+	0xbefa03ff, 0x00000400,
+	0xbeff03c1, 0xb9712a05,
+	0x80718171, 0x8f718271,
+	0x907c9973, 0x877c817c,
+	0xbf06817c, 0xbf850017,
 	0xbef603ff, 0x01000000,
-	0x877c8172, 0xbf06817c,
-	0xbefc0380, 0xbf800000,
-	0xbf85000b, 0xbe802e00,
-	0x7e000200, 0xe0704000,
-	0x7a5d0000, 0x807aff7a,
-	0x00000080, 0x807c817c,
-	0xbf0aff7c, 0x0000006a,
-	0xbf85fff6, 0xbf82000a,
-	0xbe802e00, 0x7e000200,
-	0xe0704000, 0x7a5d0000,
-	0x807aff7a, 0x00000100,
-	0x807c817c, 0xbf0aff7c,
-	0x0000006a, 0xbf85fff6,
-	0xbef60384, 0xbef603ff,
-	0x01000000, 0x877c8172,
-	0xbf06817c, 0xbf850030,
-	0x7e00027b, 0xe0704000,
-	0x7a5d0000, 0x807aff7a,
-	0x00000080, 0x7e00026c,
-	0xe0704000, 0x7a5d0000,
-	0x807aff7a, 0x00000080,
-	0x7e00026d, 0xe0704000,
-	0x7a5d0000, 0x807aff7a,
-	0x00000080, 0x7e00026e,
-	0xe0704000, 0x7a5d0000,
-	0x807aff7a, 0x00000080,
-	0x7e00026f, 0xe0704000,
-	0x7a5d0000, 0x807aff7a,
-	0x00000080, 0x7e000270,
-	0xe0704000, 0x7a5d0000,
-	0x807aff7a, 0x00000080,
-	0xb971f803, 0x7e000271,
+	0xbefc0384, 0xbf0a717c,
+	0xbf840037, 0x7e008700,
+	0x7e028701, 0x7e048702,
+	0x7e068703, 0xe0704000,
+	0x7a5d0000, 0xe0704080,
+	0x7a5d0100, 0xe0704100,
+	0x7a5d0200, 0xe0704180,
+	0x7a5d0300, 0x807c847c,
+	0x807aff7a, 0x00000200,
+	0xbf0a717c, 0xbf85ffef,
+	0xbf820025, 0xbef603ff,
+	0x01000000, 0xbefc0384,
+	0xbf0a717c, 0xbf840020,
+	0x7e008700, 0x7e028701,
+	0x7e048702, 0x7e068703,
 	0xe0704000, 0x7a5d0000,
+	0xe0704100, 0x7a5d0100,
+	0xe0704200, 0x7a5d0200,
+	0xe0704300, 0x7a5d0300,
+	0x807c847c, 0x807aff7a,
+	0x00000400, 0xbf0a717c,
+	0xbf85ffef, 0xb9711e06,
+	0x8771c171, 0xbf84000c,
+	0x8f718371, 0x80717c71,
+	0xbefe03c1, 0xbeff0380,
+	0x7e008700, 0xe0704000,
+	0x7a5d0000, 0x807c817c,
 	0x807aff7a, 0x00000080,
-	0x7e000273, 0xe0704000,
-	0x7a5d0000, 0x807aff7a,
-	0x00000080, 0xb97bf801,
-	0x7e00027b, 0xe0704000,
-	0x7a5d0000, 0x807aff7a,
-	0x00000080, 0xbf82002f,
-	0x7e00027b, 0xe0704000,
-	0x7a5d0000, 0x807aff7a,
-	0x00000100, 0x7e00026c,
-	0xe0704000, 0x7a5d0000,
-	0x807aff7a, 0x00000100,
-	0x7e00026d, 0xe0704000,
-	0x7a5d0000, 0x807aff7a,
-	0x00000100, 0x7e00026e,
-	0xe0704000, 0x7a5d0000,
-	0x807aff7a, 0x00000100,
-	0x7e00026f, 0xe0704000,
-	0x7a5d0000, 0x807aff7a,
-	0x00000100, 0x7e000270,
-	0xe0704000, 0x7a5d0000,
-	0x807aff7a, 0x00000100,
-	0xb971f803, 0x7e000271,
-	0xe0704000, 0x7a5d0000,
-	0x807aff7a, 0x00000100,
-	0x7e000273, 0xe0704000,
-	0x7a5d0000, 0x807aff7a,
-	0x00000100, 0xb97bf801,
-	0x7e00027b, 0xe0704000,
-	0x7a5d0000, 0x807aff7a,
-	0x00000100, 0xbf820119,
-	0xbef4037e, 0x8775ff7f,
-	0x0000ffff, 0x8875ff75,
-	0x00040000, 0xbef60380,
-	0xbef703ff, 0x00807fac,
-	0x8772ff7f, 0x08000000,
-	0x90728372, 0x88777277,
-	0x8772ff7f, 0x70000000,
-	0x90728172, 0x88777277,
-	0xb97902dc, 0x8879797f,
-	0xbef80380, 0xbefe03c1,
-	0x877c8179, 0xbf06817c,
+	0xbf0a717c, 0xbf85fff8,
+	0xbf820141, 0xbef4037e,
+	0x8775ff7f, 0x0000ffff,
+	0x8875ff75, 0x00040000,
+	0xbef60380, 0xbef703ff,
+	0x10807fac, 0x8772ff7f,
+	0x08000000, 0x90728372,
+	0x88777277, 0x8772ff7f,
+	0x70000000, 0x90728172,
+	0x88777277, 0xb97302dc,
+	0x8f739973, 0x8873737f,
+	0x8772ff7f, 0x04000000,
+	0xbf840036, 0xbefe03c1,
+	0x907c9973, 0x877c817c,
+	0xbf06817c, 0xbf850002,
+	0xbeff0380, 0xbf820001,
+	0xbeff03c1, 0xb96f4306,
+	0x876fc16f, 0xbf84002b,
+	0x8f6f866f, 0x8f6f826f,
+	0xbef6036f, 0xb9782a05,
+	0x80788178, 0x907c9973,
+	0x877c817c, 0xbf06817c,
+	0xbf850002, 0x8f788978,
+	0xbf820001, 0x8f788a78,
+	0xb9721e06, 0x8f728a72,
+	0x80787278, 0x8078ff78,
+	0x00000200, 0x8078ff78,
+	0x00000080, 0xbef603ff,
+	0x01000000, 0x907c9973,
+	0x877c817c, 0xbf06817c,
+	0xbefc0380, 0xbf850009,
+	0xe0310000, 0x781d0000,
+	0x807cff7c, 0x00000080,
+	0x8078ff78, 0x00000080,
+	0xbf0a6f7c, 0xbf85fff8,
+	0xbf820008, 0xe0310000,
+	0x781d0000, 0x807cff7c,
+	0x00000100, 0x8078ff78,
+	0x00000100, 0xbf0a6f7c,
+	0xbf85fff8, 0xbef80380,
+	0xbefe03c1, 0x907c9973,
+	0x877c817c, 0xbf06817c,
 	0xbf850002, 0xbeff0380,
 	0xbf820001, 0xbeff03c1,
 	0xb96f2a05, 0x806f816f,
-	0x8f6f826f, 0x877c8179,
-	0xbf06817c, 0xbf850013,
-	0x8f76876f, 0xbef603ff,
+	0x8f6f826f, 0x907c9973,
+	0x877c817c, 0xbf06817c,
+	0xbf850021, 0xbef603ff,
 	0x01000000, 0xbef20378,
-	0x8078ff78, 0x00000080,
-	0xbefc0381, 0xe0304000,
-	0x785d0000, 0xbf8c3f70,
-	0x7e008500, 0x807c817c,
-	0x8078ff78, 0x00000080,
-	0xbf0a6f7c, 0xbf85fff7,
-	0xe0304000, 0x725d0000,
-	0xbf820023, 0x8f76886f,
+	0x8078ff78, 0x00000200,
+	0xbefc0384, 0xe0304000,
+	0x785d0000, 0xe0304080,
+	0x785d0100, 0xe0304100,
+	0x785d0200, 0xe0304180,
+	0x785d0300, 0xbf8c3f70,
+	0x7e008500, 0x7e028501,
+	0x7e048502, 0x7e068503,
+	0x807c847c, 0x8078ff78,
+	0x00000200, 0xbf0a6f7c,
+	0xbf85ffee, 0xe0304000,
+	0x725d0000, 0xe0304080,
+	0x725d0100, 0xe0304100,
+	0x725d0200, 0xe0304180,
+	0x725d0300, 0xbf820031,
 	0xbef603ff, 0x01000000,
 	0xbef20378, 0x8078ff78,
-	0x00000100, 0xbefc0381,
-	0xe0304000, 0x785d0000,
-	0xbf8c3f70, 0x7e008500,
-	0x807c817c, 0x8078ff78,
-	0x00000100, 0xbf0a6f7c,
-	0xbf85fff7, 0xb96f1e06,
-	0x876fc16f, 0xbf84000e,
-	0x8f6f836f, 0x806f7c6f,
-	0xbefe03c1, 0xbeff0380,
+	0x00000400, 0xbefc0384,
 	0xe0304000, 0x785d0000,
+	0xe0304100, 0x785d0100,
+	0xe0304200, 0x785d0200,
+	0xe0304300, 0x785d0300,
 	0xbf8c3f70, 0x7e008500,
-	0x807c817c, 0x8078ff78,
-	0x00000080, 0xbf0a6f7c,
-	0xbf85fff7, 0xbeff03c1,
-	0xe0304000, 0x725d0000,
-	0x8772ff79, 0x04000000,
-	0xbf840020, 0xbefe03c1,
-	0x877c8179, 0xbf06817c,
-	0xbf850002, 0xbeff0380,
-	0xbf820001, 0xbeff03c1,
-	0xb96f4306, 0x876fc16f,
-	0xbf840016, 0x8f6f866f,
-	0x8f6f826f, 0xbef6036f,
-	0xbef603ff, 0x01000000,
-	0x877c8172, 0xbf06817c,
-	0xbefc0380, 0xbf850007,
-	0x807cff7c, 0x00000080,
-	0x8078ff78, 0x00000080,
-	0xbf0a6f7c, 0xbf85fffa,
-	0xbf820006, 0x807cff7c,
-	0x00000100, 0x8078ff78,
-	0x00000100, 0xbf0a6f7c,
-	0xbf85fffa, 0x877c8179,
-	0xbf06817c, 0xbf850003,
-	0x8f7687ff, 0x0000006a,
-	0xbf820002, 0x8f7688ff,
-	0x0000006a, 0xbef603ff,
-	0x01000000, 0x877c8179,
-	0xbf06817c, 0xbf850012,
-	0xf4211cba, 0xf0000000,
+	0x7e028501, 0x7e048502,
+	0x7e068503, 0x807c847c,
+	0x8078ff78, 0x00000400,
+	0xbf0a6f7c, 0xbf85ffee,
+	0xb96f1e06, 0x876fc16f,
+	0xbf84000e, 0x8f6f836f,
+	0x806f7c6f, 0xbefe03c1,
+	0xbeff0380, 0xe0304000,
+	0x785d0000, 0xbf8c3f70,
+	0x7e008500, 0x807c817c,
 	0x8078ff78, 0x00000080,
-	0xbefc0381, 0xf421003a,
-	0xf0000000, 0x8078ff78,
-	0x00000080, 0xbf8cc07f,
-	0xbe803000, 0xbf800000,
-	0x807c817c, 0xbf0aff7c,
-	0x0000006a, 0xbf85fff5,
-	0xbe800372, 0xbf820011,
-	0xf4211cba, 0xf0000000,
-	0x8078ff78, 0x00000100,
-	0xbefc0381, 0xf421003a,
-	0xf0000000, 0x8078ff78,
-	0x00000100, 0xbf8cc07f,
-	0xbe803000, 0xbf800000,
-	0x807c817c, 0xbf0aff7c,
-	0x0000006a, 0xbf85fff5,
-	0xbe800372, 0xbef60384,
-	0xbef603ff, 0x01000000,
-	0x877c8179, 0xbf06817c,
-	0xbf850025, 0xf4211bfa,
-	0xf0000000, 0x8078ff78,
-	0x00000080, 0xf4211b3a,
-	0xf0000000, 0x8078ff78,
-	0x00000080, 0xf4211b7a,
-	0xf0000000, 0x8078ff78,
-	0x00000080, 0xf4211eba,
-	0xf0000000, 0x8078ff78,
-	0x00000080, 0xf4211efa,
-	0xf0000000, 0x8078ff78,
-	0x00000080, 0xf4211c3a,
-	0xf0000000, 0x8078ff78,
-	0x00000080, 0xf4211c7a,
-	0xf0000000, 0x8078ff78,
-	0x00000080, 0xf4211cfa,
-	0xf0000000, 0x8078ff78,
-	0x00000080, 0xf4211e7a,
-	0xf0000000, 0x8078ff78,
-	0x00000080, 0xbf820024,
-	0xf4211bfa, 0xf0000000,
-	0x8078ff78, 0x00000100,
+	0xbf0a6f7c, 0xbf85fff7,
+	0xbeff03c1, 0xe0304000,
+	0x725d0000, 0xe0304100,
+	0x725d0100, 0xe0304200,
+	0x725d0200, 0xe0304300,
+	0x725d0300, 0xb9782a05,
+	0x80788178, 0x907c9973,
+	0x877c817c, 0xbf06817c,
+	0xbf850002, 0x8f788978,
+	0xbf820001, 0x8f788a78,
+	0xb9721e06, 0x8f728a72,
+	0x80787278, 0x8078ff78,
+	0x00000200, 0x80f8ff78,
+	0x00000050, 0xbef603ff,
+	0x01000000, 0xbefc03ff,
+	0x0000006c, 0x80f89078,
+	0xf429003a, 0xf0000000,
+	0xbf8cc07f, 0x80fc847c,
+	0xbf800000, 0xbe803100,
+	0xbe823102, 0x80f8a078,
+	0xf42d003a, 0xf0000000,
+	0xbf8cc07f, 0x80fc887c,
+	0xbf800000, 0xbe803100,
+	0xbe823102, 0xbe843104,
+	0xbe863106, 0x80f8c078,
+	0xf431003a, 0xf0000000,
+	0xbf8cc07f, 0x80fc907c,
+	0xbf800000, 0xbe803100,
+	0xbe823102, 0xbe843104,
+	0xbe863106, 0xbe883108,
+	0xbe8a310a, 0xbe8c310c,
+	0xbe8e310e, 0xbf06807c,
+	0xbf84fff0, 0xb9782a05,
+	0x80788178, 0x907c9973,
+	0x877c817c, 0xbf06817c,
+	0xbf850002, 0x8f788978,
+	0xbf820001, 0x8f788a78,
+	0xb9721e06, 0x8f728a72,
+	0x80787278, 0x8078ff78,
+	0x00000200, 0xbef603ff,
+	0x01000000, 0xf4211bfa,
+	0xf0000000, 0x80788478,
 	0xf4211b3a, 0xf0000000,
-	0x8078ff78, 0x00000100,
-	0xf4211b7a, 0xf0000000,
-	0x8078ff78, 0x00000100,
+	0x80788478, 0xf4211b7a,
+	0xf0000000, 0x80788478,
 	0xf4211eba, 0xf0000000,
-	0x8078ff78, 0x00000100,
-	0xf4211efa, 0xf0000000,
-	0x8078ff78, 0x00000100,
+	0x80788478, 0xf4211efa,
+	0xf0000000, 0x80788478,
 	0xf4211c3a, 0xf0000000,
-	0x8078ff78, 0x00000100,
-	0xf4211c7a, 0xf0000000,
-	0x8078ff78, 0x00000100,
-	0xf4211cfa, 0xf0000000,
-	0x8078ff78, 0x00000100,
+	0x80788478, 0xf4211c7a,
+	0xf0000000, 0x80788478,
 	0xf4211e7a, 0xf0000000,
-	0x8078ff78, 0x00000100,
-	0xbf8cc07f, 0x876dff6d,
+	0x80788478, 0xf4211cfa,
+	0xf0000000, 0x80788478,
+	0xf4211bba, 0xf0000000,
+	0x80788478, 0xbf8cc07f,
+	0xb9eef814, 0xf4211bba,
+	0xf0000000, 0x80788478,
+	0xbf8cc07f, 0xb9eef815,
+	0xbef2036d, 0x876dff72,
 	0x0000ffff, 0xbefc036f,
 	0xbefe037a, 0xbeff037b,
 	0x876f71ff, 0x000003ff,
-	0xb9ef4803, 0xb9f3f816,
+	0xb9ef4803, 0xb9f9f816,
 	0x876f71ff, 0xfffff800,
 	0x906f8b6f, 0xb9efa2c3,
-	0xb9f9f801, 0x876fff6d,
-	0xf0000000, 0x906f9c6f,
-	0x8f6f906f, 0xbef20380,
-	0x88726f72, 0x876fff6d,
-	0x08000000, 0x906f9b6f,
-	0x8f6f8f6f, 0x88726f72,
-	0x876fff70, 0x00800000,
-	0x906f976f, 0xb9f2f807,
-	0xb9f0f802, 0xbf8a0000,
-	0xbe80226c, 0xbf810000,
+	0xb9f3f801, 0x876fff72,
+	0xfc000000, 0x906f9a6f,
+	0x8f6f906f, 0xbef30380,
+	0x88736f73, 0x876fff72,
+	0x02000000, 0x906f996f,
+	0x8f6f8f6f, 0x88736f73,
+	0x876fff72, 0x01000000,
+	0x906f986f, 0x8f6f996f,
+	0x88736f73, 0x876fff70,
+	0x00800000, 0x906f976f,
+	0xb9f3f807, 0x87fe7e7e,
+	0x87ea6a6a, 0xb9f0f802,
+	0xbf8a0000, 0xbe80226c,
+	0xbf810000, 0xbf9f0000,
 	0xbf9f0000, 0xbf9f0000,
 	0xbf9f0000, 0xbf9f0000,
-	0xbf9f0000, 0x00000000,
+};
+static const uint32_t cwsr_trap_arcturus_hex[] = {
+	0xbf820001, 0xbf8202c4,
+	0xb8f8f802, 0x89788678,
+	0xb8eef801, 0x866eff6e,
+	0x00000800, 0xbf840003,
+	0x866eff78, 0x00002000,
+	0xbf840016, 0xb8fbf803,
+	0x866eff7b, 0x00000400,
+	0xbf85003b, 0x866eff7b,
+	0x00000800, 0xbf850003,
+	0x866eff7b, 0x00000100,
+	0xbf84000c, 0x866eff78,
+	0x00002000, 0xbf840005,
+	0xbf8e0010, 0xb8eef803,
+	0x866eff6e, 0x00000400,
+	0xbf84fffb, 0x8778ff78,
+	0x00002000, 0x80ec886c,
+	0x82ed806d, 0xb8eef807,
+	0x866fff6e, 0x001f8000,
+	0x8e6f8b6f, 0x8977ff77,
+	0xfc000000, 0x87776f77,
+	0x896eff6e, 0x001f8000,
+	0xb96ef807, 0xb8faf812,
+	0xb8fbf813, 0x8efa887a,
+	0xc0071bbd, 0x00000000,
+	0xbf8cc07f, 0xc0071ebd,
+	0x00000008, 0xbf8cc07f,
+	0x86ee6e6e, 0xbf840001,
+	0xbe801d6e, 0xb8fbf803,
+	0x867bff7b, 0x000001ff,
+	0xbf850002, 0x806c846c,
+	0x826d806d, 0x866dff6d,
+	0x0000ffff, 0x8f6e8b77,
+	0x866eff6e, 0x001f8000,
+	0xb96ef807, 0x86fe7e7e,
+	0x86ea6a6a, 0x8f6e8378,
+	0xb96ee0c2, 0xbf800002,
+	0xb9780002, 0xbe801f6c,
+	0x866dff6d, 0x0000ffff,
+	0xbefa0080, 0xb97a0283,
+	0xb8fa2407, 0x8e7a9b7a,
+	0x876d7a6d, 0xb8fa03c7,
+	0x8e7a9a7a, 0x876d7a6d,
+	0xb8faf807, 0x867aff7a,
+	0x00007fff, 0xb97af807,
+	0xbeee007e, 0xbeef007f,
+	0xbefe0180, 0xbf900004,
+	0x877a8478, 0xb97af802,
+	0xbf8e0002, 0xbf88fffe,
+	0xb8fa2a05, 0x807a817a,
+	0x8e7a8a7a, 0x8e7a817a,
+	0xb8fb1605, 0x807b817b,
+	0x8e7b867b, 0x807a7b7a,
+	0x807a7e7a, 0x827b807f,
+	0x867bff7b, 0x0000ffff,
+	0xc04b1c3d, 0x00000050,
+	0xbf8cc07f, 0xc04b1d3d,
+	0x00000060, 0xbf8cc07f,
+	0xc0431e7d, 0x00000074,
+	0xbf8cc07f, 0xbef4007e,
+	0x8675ff7f, 0x0000ffff,
+	0x8775ff75, 0x00040000,
+	0xbef60080, 0xbef700ff,
+	0x00807fac, 0x867aff7f,
+	0x08000000, 0x8f7a837a,
+	0x87777a77, 0x867aff7f,
+	0x70000000, 0x8f7a817a,
+	0x87777a77, 0xbef1007c,
+	0xbef00080, 0xb8f02a05,
+	0x80708170, 0x8e708a70,
+	0x8e708170, 0xb8fa1605,
+	0x807a817a, 0x8e7a867a,
+	0x80707a70, 0xbef60084,
+	0xbef600ff, 0x01000000,
+	0xbefe007c, 0xbefc0070,
+	0xc0611c7a, 0x0000007c,
+	0xbf8cc07f, 0x80708470,
+	0xbefc007e, 0xbefe007c,
+	0xbefc0070, 0xc0611b3a,
+	0x0000007c, 0xbf8cc07f,
+	0x80708470, 0xbefc007e,
+	0xbefe007c, 0xbefc0070,
+	0xc0611b7a, 0x0000007c,
+	0xbf8cc07f, 0x80708470,
+	0xbefc007e, 0xbefe007c,
+	0xbefc0070, 0xc0611bba,
+	0x0000007c, 0xbf8cc07f,
+	0x80708470, 0xbefc007e,
+	0xbefe007c, 0xbefc0070,
+	0xc0611bfa, 0x0000007c,
+	0xbf8cc07f, 0x80708470,
+	0xbefc007e, 0xbefe007c,
+	0xbefc0070, 0xc0611e3a,
+	0x0000007c, 0xbf8cc07f,
+	0x80708470, 0xbefc007e,
+	0xb8fbf803, 0xbefe007c,
+	0xbefc0070, 0xc0611efa,
+	0x0000007c, 0xbf8cc07f,
+	0x80708470, 0xbefc007e,
+	0xbefe007c, 0xbefc0070,
+	0xc0611a3a, 0x0000007c,
+	0xbf8cc07f, 0x80708470,
+	0xbefc007e, 0xbefe007c,
+	0xbefc0070, 0xc0611a7a,
+	0x0000007c, 0xbf8cc07f,
+	0x80708470, 0xbefc007e,
+	0xb8f1f801, 0xbefe007c,
+	0xbefc0070, 0xc0611c7a,
+	0x0000007c, 0xbf8cc07f,
+	0x80708470, 0xbefc007e,
+	0x867aff7f, 0x04000000,
+	0xbeef0080, 0x876f6f7a,
+	0xb8f02a05, 0x80708170,
+	0x8e708a70, 0x8e708170,
+	0xb8fb1605, 0x807b817b,
+	0x8e7b847b, 0x8e76827b,
+	0xbef600ff, 0x01000000,
+	0xbef20174, 0x80747074,
+	0x82758075, 0xbefc0080,
+	0xbf800000, 0xbe802b00,
+	0xbe822b02, 0xbe842b04,
+	0xbe862b06, 0xbe882b08,
+	0xbe8a2b0a, 0xbe8c2b0c,
+	0xbe8e2b0e, 0xc06b003a,
+	0x00000000, 0xbf8cc07f,
+	0xc06b013a, 0x00000010,
+	0xbf8cc07f, 0xc06b023a,
+	0x00000020, 0xbf8cc07f,
+	0xc06b033a, 0x00000030,
+	0xbf8cc07f, 0x8074c074,
+	0x82758075, 0x807c907c,
+	0xbf0a7b7c, 0xbf85ffe7,
+	0xbef40172, 0xbef00080,
+	0xbefe00c1, 0xbeff00c1,
+	0xbee80080, 0xbee90080,
+	0xbef600ff, 0x01000000,
+	0x867aff78, 0x00400000,
+	0xbf850003, 0xb8faf803,
+	0x897a7aff, 0x10000000,
+	0xbf85004d, 0xbe840080,
+	0xd2890000, 0x00000900,
+	0x80048104, 0xd2890001,
+	0x00000900, 0x80048104,
+	0xd2890002, 0x00000900,
+	0x80048104, 0xd2890003,
+	0x00000900, 0x80048104,
+	0xc069003a, 0x00000070,
+	0xbf8cc07f, 0x80709070,
+	0xbf06c004, 0xbf84ffee,
+	0xbe840080, 0xd2890000,
+	0x00000901, 0x80048104,
+	0xd2890001, 0x00000901,
+	0x80048104, 0xd2890002,
+	0x00000901, 0x80048104,
+	0xd2890003, 0x00000901,
+	0x80048104, 0xc069003a,
+	0x00000070, 0xbf8cc07f,
+	0x80709070, 0xbf06c004,
+	0xbf84ffee, 0xbe840080,
+	0xd2890000, 0x00000902,
+	0x80048104, 0xd2890001,
+	0x00000902, 0x80048104,
+	0xd2890002, 0x00000902,
+	0x80048104, 0xd2890003,
+	0x00000902, 0x80048104,
+	0xc069003a, 0x00000070,
+	0xbf8cc07f, 0x80709070,
+	0xbf06c004, 0xbf84ffee,
+	0xbe840080, 0xd2890000,
+	0x00000903, 0x80048104,
+	0xd2890001, 0x00000903,
+	0x80048104, 0xd2890002,
+	0x00000903, 0x80048104,
+	0xd2890003, 0x00000903,
+	0x80048104, 0xc069003a,
+	0x00000070, 0xbf8cc07f,
+	0x80709070, 0xbf06c004,
+	0xbf84ffee, 0xbf820008,
+	0xe0724000, 0x701d0000,
+	0xe0724100, 0x701d0100,
+	0xe0724200, 0x701d0200,
+	0xe0724300, 0x701d0300,
+	0xbefe00c1, 0xbeff00c1,
+	0xb8fb4306, 0x867bc17b,
+	0xbf840064, 0xbf8a0000,
+	0x867aff6f, 0x04000000,
+	0xbf840060, 0x8e7b867b,
+	0x8e7b827b, 0xbef6007b,
+	0xb8f02a05, 0x80708170,
+	0x8e708a70, 0x8e708170,
+	0xb8fa1605, 0x807a817a,
+	0x8e7a867a, 0x80707a70,
+	0x8070ff70, 0x00000080,
+	0xbef600ff, 0x01000000,
+	0xbefc0080, 0xd28c0002,
+	0x000100c1, 0xd28d0003,
+	0x000204c1, 0x867aff78,
+	0x00400000, 0xbf850003,
+	0xb8faf803, 0x897a7aff,
+	0x10000000, 0xbf850030,
+	0x24040682, 0xd86e4000,
+	0x00000002, 0xbf8cc07f,
+	0xbe840080, 0xd2890000,
+	0x00000900, 0x80048104,
+	0xd2890001, 0x00000900,
+	0x80048104, 0xd2890002,
+	0x00000900, 0x80048104,
+	0xd2890003, 0x00000900,
+	0x80048104, 0xc069003a,
+	0x00000070, 0xbf8cc07f,
+	0x80709070, 0xbf06c004,
+	0xbf84ffee, 0xbe840080,
+	0xd2890000, 0x00000901,
+	0x80048104, 0xd2890001,
+	0x00000901, 0x80048104,
+	0xd2890002, 0x00000901,
+	0x80048104, 0xd2890003,
+	0x00000901, 0x80048104,
+	0xc069003a, 0x00000070,
+	0xbf8cc07f, 0x80709070,
+	0xbf06c004, 0xbf84ffee,
+	0x680404ff, 0x00000200,
+	0xd0c9006a, 0x0000f702,
+	0xbf87ffd2, 0xbf820015,
+	0xd1060002, 0x00011103,
+	0x7e0602ff, 0x00000200,
+	0xbefc00ff, 0x00010000,
+	0xbe800077, 0x8677ff77,
+	0xff7fffff, 0x8777ff77,
+	0x00058000, 0xd8ec0000,
+	0x00000002, 0xbf8cc07f,
+	0xe0765000, 0x701d0002,
+	0x68040702, 0xd0c9006a,
+	0x0000f702, 0xbf87fff7,
+	0xbef70000, 0xbef000ff,
+	0x00000400, 0xbefe00c1,
+	0xbeff00c1, 0xb8fb2a05,
+	0x807b817b, 0x8e7b827b,
+	0x8e76887b, 0xbef600ff,
+	0x01000000, 0xbefc0084,
+	0xbf0a7b7c, 0xbf84006d,
+	0xbf11017c, 0x807bff7b,
+	0x00001000, 0x867aff78,
+	0x00400000, 0xbf850003,
+	0xb8faf803, 0x897a7aff,
+	0x10000000, 0xbf850051,
+	0xbe840080, 0xd2890000,
+	0x00000900, 0x80048104,
+	0xd2890001, 0x00000900,
+	0x80048104, 0xd2890002,
+	0x00000900, 0x80048104,
+	0xd2890003, 0x00000900,
+	0x80048104, 0xc069003a,
+	0x00000070, 0xbf8cc07f,
+	0x80709070, 0xbf06c004,
+	0xbf84ffee, 0xbe840080,
+	0xd2890000, 0x00000901,
+	0x80048104, 0xd2890001,
+	0x00000901, 0x80048104,
+	0xd2890002, 0x00000901,
+	0x80048104, 0xd2890003,
+	0x00000901, 0x80048104,
+	0xc069003a, 0x00000070,
+	0xbf8cc07f, 0x80709070,
+	0xbf06c004, 0xbf84ffee,
+	0xbe840080, 0xd2890000,
+	0x00000902, 0x80048104,
+	0xd2890001, 0x00000902,
+	0x80048104, 0xd2890002,
+	0x00000902, 0x80048104,
+	0xd2890003, 0x00000902,
+	0x80048104, 0xc069003a,
+	0x00000070, 0xbf8cc07f,
+	0x80709070, 0xbf06c004,
+	0xbf84ffee, 0xbe840080,
+	0xd2890000, 0x00000903,
+	0x80048104, 0xd2890001,
+	0x00000903, 0x80048104,
+	0xd2890002, 0x00000903,
+	0x80048104, 0xd2890003,
+	0x00000903, 0x80048104,
+	0xc069003a, 0x00000070,
+	0xbf8cc07f, 0x80709070,
+	0xbf06c004, 0xbf84ffee,
+	0x807c847c, 0xbf0a7b7c,
+	0xbf85ffb1, 0xbf9c0000,
+	0xbf820012, 0x7e000300,
+	0x7e020301, 0x7e040302,
+	0x7e060303, 0xe0724000,
+	0x701d0000, 0xe0724100,
+	0x701d0100, 0xe0724200,
+	0x701d0200, 0xe0724300,
+	0x701d0300, 0x807c847c,
+	0x8070ff70, 0x00000400,
+	0xbf0a7b7c, 0xbf85ffef,
+	0xbf9c0000, 0xbefc0080,
+	0xbf11017c, 0x867aff78,
+	0x00400000, 0xbf850003,
+	0xb8faf803, 0x897a7aff,
+	0x10000000, 0xbf850059,
+	0xd3d84000, 0x18000100,
+	0xd3d84001, 0x18000101,
+	0xd3d84002, 0x18000102,
+	0xd3d84003, 0x18000103,
+	0xbe840080, 0xd2890000,
+	0x00000900, 0x80048104,
+	0xd2890001, 0x00000900,
+	0x80048104, 0xd2890002,
+	0x00000900, 0x80048104,
+	0xd2890003, 0x00000900,
+	0x80048104, 0xc069003a,
+	0x00000070, 0xbf8cc07f,
+	0x80709070, 0xbf06c004,
+	0xbf84ffee, 0xbe840080,
+	0xd2890000, 0x00000901,
+	0x80048104, 0xd2890001,
+	0x00000901, 0x80048104,
+	0xd2890002, 0x00000901,
+	0x80048104, 0xd2890003,
+	0x00000901, 0x80048104,
+	0xc069003a, 0x00000070,
+	0xbf8cc07f, 0x80709070,
+	0xbf06c004, 0xbf84ffee,
+	0xbe840080, 0xd2890000,
+	0x00000902, 0x80048104,
+	0xd2890001, 0x00000902,
+	0x80048104, 0xd2890002,
+	0x00000902, 0x80048104,
+	0xd2890003, 0x00000902,
+	0x80048104, 0xc069003a,
+	0x00000070, 0xbf8cc07f,
+	0x80709070, 0xbf06c004,
+	0xbf84ffee, 0xbe840080,
+	0xd2890000, 0x00000903,
+	0x80048104, 0xd2890001,
+	0x00000903, 0x80048104,
+	0xd2890002, 0x00000903,
+	0x80048104, 0xd2890003,
+	0x00000903, 0x80048104,
+	0xc069003a, 0x00000070,
+	0xbf8cc07f, 0x80709070,
+	0xbf06c004, 0xbf84ffee,
+	0x807c847c, 0xbf0a7b7c,
+	0xbf85ffa9, 0xbf9c0000,
+	0xbf820016, 0xd3d84000,
+	0x18000100, 0xd3d84001,
+	0x18000101, 0xd3d84002,
+	0x18000102, 0xd3d84003,
+	0x18000103, 0xe0724000,
+	0x701d0000, 0xe0724100,
+	0x701d0100, 0xe0724200,
+	0x701d0200, 0xe0724300,
+	0x701d0300, 0x807c847c,
+	0x8070ff70, 0x00000400,
+	0xbf0a7b7c, 0xbf85ffeb,
+	0xbf9c0000, 0xbf820106,
+	0xbef4007e, 0x8675ff7f,
+	0x0000ffff, 0x8775ff75,
+	0x00040000, 0xbef60080,
+	0xbef700ff, 0x00807fac,
+	0x866eff7f, 0x08000000,
+	0x8f6e836e, 0x87776e77,
+	0x866eff7f, 0x70000000,
+	0x8f6e816e, 0x87776e77,
+	0x866eff7f, 0x04000000,
+	0xbf84001f, 0xbefe00c1,
+	0xbeff00c1, 0xb8ef4306,
+	0x866fc16f, 0xbf84001a,
+	0x8e6f866f, 0x8e6f826f,
+	0xbef6006f, 0xb8f82a05,
+	0x80788178, 0x8e788a78,
+	0x8e788178, 0xb8ee1605,
+	0x806e816e, 0x8e6e866e,
+	0x80786e78, 0x8078ff78,
+	0x00000080, 0xbef600ff,
+	0x01000000, 0xbefc0080,
+	0xe0510000, 0x781d0000,
+	0xe0510100, 0x781d0000,
+	0x807cff7c, 0x00000200,
+	0x8078ff78, 0x00000200,
+	0xbf0a6f7c, 0xbf85fff6,
+	0xbef80080, 0xbefe00c1,
+	0xbeff00c1, 0xb8ef2a05,
+	0x806f816f, 0x8e6f826f,
+	0x8e76886f, 0xbef90076,
+	0xbef600ff, 0x01000000,
+	0xbeee0078, 0x8078ff78,
+	0x00000400, 0xbef30079,
+	0x8079ff79, 0x00000400,
+	0xbefc0084, 0xbf11087c,
+	0x806fff6f, 0x00008000,
+	0xe0524000, 0x791d0000,
+	0xe0524100, 0x791d0100,
+	0xe0524200, 0x791d0200,
+	0xe0524300, 0x791d0300,
+	0x8079ff79, 0x00000400,
+	0xbf8c0f70, 0xd3d94000,
+	0x18000100, 0xd3d94001,
+	0x18000101, 0xd3d94002,
+	0x18000102, 0xd3d94003,
+	0x18000103, 0xe0524000,
+	0x781d0000, 0xe0524100,
+	0x781d0100, 0xe0524200,
+	0x781d0200, 0xe0524300,
+	0x781d0300, 0xbf8c0f70,
+	0x7e000300, 0x7e020301,
+	0x7e040302, 0x7e060303,
+	0x807c847c, 0x8078ff78,
+	0x00000400, 0xbf0a6f7c,
+	0xbf85ffdb, 0xbf9c0000,
+	0xe0524000, 0x731d0000,
+	0xe0524100, 0x731d0100,
+	0xe0524200, 0x731d0200,
+	0xe0524300, 0x731d0300,
+	0xbf8c0f70, 0xd3d94000,
+	0x18000100, 0xd3d94001,
+	0x18000101, 0xd3d94002,
+	0x18000102, 0xd3d94003,
+	0x18000103, 0xe0524000,
+	0x6e1d0000, 0xe0524100,
+	0x6e1d0100, 0xe0524200,
+	0x6e1d0200, 0xe0524300,
+	0x6e1d0300, 0xb8f82a05,
+	0x80788178, 0x8e788a78,
+	0x8e788178, 0xb8ee1605,
+	0x806e816e, 0x8e6e866e,
+	0x80786e78, 0x80f8c078,
+	0xb8ef1605, 0x806f816f,
+	0x8e6f846f, 0x8e76826f,
+	0xbef600ff, 0x01000000,
+	0xbefc006f, 0xc031003a,
+	0x00000078, 0x80f8c078,
+	0xbf8cc07f, 0x80fc907c,
+	0xbf800000, 0xbe802d00,
+	0xbe822d02, 0xbe842d04,
+	0xbe862d06, 0xbe882d08,
+	0xbe8a2d0a, 0xbe8c2d0c,
+	0xbe8e2d0e, 0xbf06807c,
+	0xbf84fff0, 0xb8f82a05,
+	0x80788178, 0x8e788a78,
+	0x8e788178, 0xb8ee1605,
+	0x806e816e, 0x8e6e866e,
+	0x80786e78, 0xbef60084,
+	0xbef600ff, 0x01000000,
+	0xc0211bfa, 0x00000078,
+	0x80788478, 0xc0211b3a,
+	0x00000078, 0x80788478,
+	0xc0211b7a, 0x00000078,
+	0x80788478, 0xc0211c3a,
+	0x00000078, 0x80788478,
+	0xc0211c7a, 0x00000078,
+	0x80788478, 0xc0211eba,
+	0x00000078, 0x80788478,
+	0xc0211efa, 0x00000078,
+	0x80788478, 0xc0211a3a,
+	0x00000078, 0x80788478,
+	0xc0211a7a, 0x00000078,
+	0x80788478, 0xc0211cfa,
+	0x00000078, 0x80788478,
+	0xbf8cc07f, 0xbefc006f,
+	0xbefe0070, 0xbeff0071,
+	0x866f7bff, 0x000003ff,
+	0xb96f4803, 0x866f7bff,
+	0xfffff800, 0x8f6f8b6f,
+	0xb96fa2c3, 0xb973f801,
+	0xb8ee2a05, 0x806e816e,
+	0x8e6e8a6e, 0x8e6e816e,
+	0xb8ef1605, 0x806f816f,
+	0x8e6f866f, 0x806e6f6e,
+	0x806e746e, 0x826f8075,
+	0x866fff6f, 0x0000ffff,
+	0xc00b1c37, 0x00000050,
+	0xc00b1d37, 0x00000060,
+	0xc0031e77, 0x00000074,
+	0xbf8cc07f, 0x866fff6d,
+	0xf8000000, 0x8f6f9b6f,
+	0x8e6f906f, 0xbeee0080,
+	0x876e6f6e, 0x866fff6d,
+	0x04000000, 0x8f6f9a6f,
+	0x8e6f8f6f, 0x876e6f6e,
+	0x866fff7a, 0x00800000,
+	0x8f6f976f, 0xb96ef807,
+	0x866dff6d, 0x0000ffff,
+	0x86fe7e7e, 0x86ea6a6a,
+	0x8f6e837a, 0xb96ee0c2,
+	0xbf800002, 0xb97a0002,
+	0xbf8a0000, 0x95806f6c,
+	0xbf810000, 0x00000000,
 };
diff --git a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx10.asm b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx10.asm
index f20e463e748b..35986219ce5f 100644
--- a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx10.asm
+++ b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx10.asm
@@ -20,1105 +20,947 @@
  * OTHER DEALINGS IN THE SOFTWARE.
  */
 
+var SQ_WAVE_STATUS_INST_ATC_SHIFT		= 23
+var SQ_WAVE_STATUS_INST_ATC_MASK		= 0x00800000
+var SQ_WAVE_STATUS_SPI_PRIO_MASK		= 0x00000006
+var SQ_WAVE_STATUS_HALT_MASK			= 0x2000
+
+var SQ_WAVE_LDS_ALLOC_LDS_SIZE_SHIFT		= 12
+var SQ_WAVE_LDS_ALLOC_LDS_SIZE_SIZE		= 9
+var SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SHIFT		= 8
+var SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SIZE		= 6
+var SQ_WAVE_GPR_ALLOC_SGPR_SIZE_SHIFT		= 24
+var SQ_WAVE_GPR_ALLOC_SGPR_SIZE_SIZE		= 4
+var SQ_WAVE_LDS_ALLOC_VGPR_SHARED_SIZE_SHIFT	= 24
+var SQ_WAVE_LDS_ALLOC_VGPR_SHARED_SIZE_SIZE	= 4
+var SQ_WAVE_IB_STS2_WAVE64_SHIFT		= 11
+var SQ_WAVE_IB_STS2_WAVE64_SIZE			= 1
+
+var SQ_WAVE_TRAPSTS_SAVECTX_MASK		= 0x400
+var SQ_WAVE_TRAPSTS_EXCE_MASK			= 0x1FF
+var SQ_WAVE_TRAPSTS_SAVECTX_SHIFT		= 10
+var SQ_WAVE_TRAPSTS_MEM_VIOL_MASK		= 0x100
+var SQ_WAVE_TRAPSTS_MEM_VIOL_SHIFT		= 8
+var SQ_WAVE_TRAPSTS_PRE_SAVECTX_MASK		= 0x3FF
+var SQ_WAVE_TRAPSTS_PRE_SAVECTX_SHIFT		= 0x0
+var SQ_WAVE_TRAPSTS_PRE_SAVECTX_SIZE		= 10
+var SQ_WAVE_TRAPSTS_POST_SAVECTX_MASK		= 0xFFFFF800
+var SQ_WAVE_TRAPSTS_POST_SAVECTX_SHIFT		= 11
+var SQ_WAVE_TRAPSTS_POST_SAVECTX_SIZE		= 21
+var SQ_WAVE_TRAPSTS_ILLEGAL_INST_MASK		= 0x800
+
+var SQ_WAVE_IB_STS_RCNT_SHIFT			= 16
+var SQ_WAVE_IB_STS_FIRST_REPLAY_SHIFT		= 15
+var SQ_WAVE_IB_STS_REPLAY_W64H_SHIFT		= 25
+var SQ_WAVE_IB_STS_REPLAY_W64H_SIZE		= 1
+var SQ_WAVE_IB_STS_REPLAY_W64H_MASK		= 0x02000000
+var SQ_WAVE_IB_STS_FIRST_REPLAY_SIZE		= 1
+var SQ_WAVE_IB_STS_RCNT_SIZE			= 6
+var SQ_WAVE_IB_STS_RCNT_FIRST_REPLAY_MASK	= 0x003F8000
+var SQ_WAVE_IB_STS_RCNT_FIRST_REPLAY_MASK_NEG	= 0x00007FFF
+
+var SQ_BUF_RSRC_WORD1_ATC_SHIFT			= 24
+var SQ_BUF_RSRC_WORD3_MTYPE_SHIFT		= 27
+
+// bits [31:24] unused by SPI debug data
+var TTMP11_SAVE_REPLAY_W64H_SHIFT		= 31
+var TTMP11_SAVE_REPLAY_W64H_MASK		= 0x80000000
+var TTMP11_SAVE_RCNT_FIRST_REPLAY_SHIFT		= 24
+var TTMP11_SAVE_RCNT_FIRST_REPLAY_MASK		= 0x7F000000
+
+// SQ_SEL_X/Y/Z/W, BUF_NUM_FORMAT_FLOAT, (0 for MUBUF stride[17:14]
+// when ADD_TID_ENABLE and BUF_DATA_FORMAT_32 for MTBUF), ADD_TID_ENABLE
+var S_SAVE_BUF_RSRC_WORD1_STRIDE		= 0x00040000
+var S_SAVE_BUF_RSRC_WORD3_MISC			= 0x10807FAC
+
+var S_SAVE_SPI_INIT_ATC_MASK			= 0x08000000
+var S_SAVE_SPI_INIT_ATC_SHIFT			= 27
+var S_SAVE_SPI_INIT_MTYPE_MASK			= 0x70000000
+var S_SAVE_SPI_INIT_MTYPE_SHIFT			= 28
+var S_SAVE_SPI_INIT_FIRST_WAVE_MASK		= 0x04000000
+var S_SAVE_SPI_INIT_FIRST_WAVE_SHIFT		= 26
+
+var S_SAVE_PC_HI_RCNT_SHIFT			= 26
+var S_SAVE_PC_HI_RCNT_MASK			= 0xFC000000
+var S_SAVE_PC_HI_FIRST_REPLAY_SHIFT		= 25
+var S_SAVE_PC_HI_FIRST_REPLAY_MASK		= 0x02000000
+var S_SAVE_PC_HI_REPLAY_W64H_SHIFT		= 24
+var S_SAVE_PC_HI_REPLAY_W64H_MASK		= 0x01000000
+
+var s_sgpr_save_num				= 108
+
+var s_save_spi_init_lo				= exec_lo
+var s_save_spi_init_hi				= exec_hi
+var s_save_pc_lo				= ttmp0
+var s_save_pc_hi				= ttmp1
+var s_save_exec_lo				= ttmp2
+var s_save_exec_hi				= ttmp3
+var s_save_status				= ttmp12
+var s_save_trapsts				= ttmp5
+var s_save_xnack_mask				= ttmp6
+var s_wave_size					= ttmp7
+var s_save_buf_rsrc0				= ttmp8
+var s_save_buf_rsrc1				= ttmp9
+var s_save_buf_rsrc2				= ttmp10
+var s_save_buf_rsrc3				= ttmp11
+var s_save_mem_offset				= ttmp14
+var s_save_alloc_size				= s_save_trapsts
+var s_save_tmp					= s_save_buf_rsrc2
+var s_save_m0					= ttmp15
+
+var S_RESTORE_BUF_RSRC_WORD1_STRIDE		= S_SAVE_BUF_RSRC_WORD1_STRIDE
+var S_RESTORE_BUF_RSRC_WORD3_MISC		= S_SAVE_BUF_RSRC_WORD3_MISC
+
+var S_RESTORE_SPI_INIT_ATC_MASK			= 0x08000000
+var S_RESTORE_SPI_INIT_ATC_SHIFT		= 27
+var S_RESTORE_SPI_INIT_MTYPE_MASK		= 0x70000000
+var S_RESTORE_SPI_INIT_MTYPE_SHIFT		= 28
+var S_RESTORE_SPI_INIT_FIRST_WAVE_MASK		= 0x04000000
+var S_RESTORE_SPI_INIT_FIRST_WAVE_SHIFT		= 26
+var S_WAVE_SIZE					= 25
+
+var S_RESTORE_PC_HI_RCNT_SHIFT			= S_SAVE_PC_HI_RCNT_SHIFT
+var S_RESTORE_PC_HI_RCNT_MASK			= S_SAVE_PC_HI_RCNT_MASK
+var S_RESTORE_PC_HI_FIRST_REPLAY_SHIFT		= S_SAVE_PC_HI_FIRST_REPLAY_SHIFT
+var S_RESTORE_PC_HI_FIRST_REPLAY_MASK		= S_SAVE_PC_HI_FIRST_REPLAY_MASK
+
+var s_restore_spi_init_lo			= exec_lo
+var s_restore_spi_init_hi			= exec_hi
+var s_restore_mem_offset			= ttmp12
+var s_restore_alloc_size			= ttmp3
+var s_restore_tmp				= ttmp6
+var s_restore_mem_offset_save			= s_restore_tmp
+var s_restore_m0				= s_restore_alloc_size
+var s_restore_mode				= ttmp7
+var s_restore_flat_scratch			= ttmp2
+var s_restore_pc_lo				= ttmp0
+var s_restore_pc_hi				= ttmp1
+var s_restore_exec_lo				= ttmp14
+var s_restore_exec_hi				= ttmp15
+var s_restore_status				= ttmp4
+var s_restore_trapsts				= ttmp5
+var s_restore_xnack_mask			= ttmp13
+var s_restore_buf_rsrc0				= ttmp8
+var s_restore_buf_rsrc1				= ttmp9
+var s_restore_buf_rsrc2				= ttmp10
+var s_restore_buf_rsrc3				= ttmp11
+var s_restore_size				= ttmp7
 
 shader main
+	asic(DEFAULT)
+	type(CS)
+	wave_size(32)
 
-asic(DEFAULT)
-
-type(CS)
-
-wave_size(32)
-/*************************************************************************/
-/*					control on how to run the shader					 */
-/*************************************************************************/
-//any hack that needs to be made to run this code in EMU (either becasue various EMU code are not ready or no compute save & restore in EMU run)
-var EMU_RUN_HACK					=	0
-var EMU_RUN_HACK_RESTORE_NORMAL		=	0
-var EMU_RUN_HACK_SAVE_NORMAL_EXIT	=	0
-var	EMU_RUN_HACK_SAVE_SINGLE_WAVE	=	0
-var EMU_RUN_HACK_SAVE_FIRST_TIME	= 	0					//for interrupted restore in which the first save is through EMU_RUN_HACK
-var SAVE_LDS						= 	0
-var WG_BASE_ADDR_LO					=   0x9000a000
-var WG_BASE_ADDR_HI					=	0x0
-var WAVE_SPACE						=	0x9000				//memory size that each wave occupies in workgroup state mem, increase from 5000 to 9000 for more SGPR need to be saved
-var CTX_SAVE_CONTROL				=	0x0
-var CTX_RESTORE_CONTROL				=	CTX_SAVE_CONTROL
-var SIM_RUN_HACK					=	0					//any hack that needs to be made to run this code in SIM (either becasue various RTL code are not ready or no compute save & restore in RTL run)
-var	SGPR_SAVE_USE_SQC				=	0					//use SQC D$ to do the write
-var USE_MTBUF_INSTEAD_OF_MUBUF		=	0					//need to change BUF_DATA_FORMAT in S_SAVE_BUF_RSRC_WORD3_MISC from 0 to BUF_DATA_FORMAT_32 if set to 1 (i.e. 0x00827FAC)
-var SWIZZLE_EN						=	0					//whether we use swizzled buffer addressing
-var SAVE_RESTORE_HWID_DDID          =   0
-var RESTORE_DDID_IN_SGPR18          =   0
-/**************************************************************************/
-/*                     	variables							              */
-/**************************************************************************/
-var SQ_WAVE_STATUS_INST_ATC_SHIFT  = 23
-var SQ_WAVE_STATUS_INST_ATC_MASK   = 0x00800000
-var SQ_WAVE_STATUS_SPI_PRIO_MASK   = 0x00000006
-
-var SQ_WAVE_LDS_ALLOC_LDS_SIZE_SHIFT	= 12
-var SQ_WAVE_LDS_ALLOC_LDS_SIZE_SIZE		= 9
-var SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SHIFT	= 8
-var SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SIZE	= 6
-var SQ_WAVE_GPR_ALLOC_SGPR_SIZE_SHIFT	= 24
-var SQ_WAVE_GPR_ALLOC_SGPR_SIZE_SIZE	= 4						//FIXME	 sq.blk still has 4 bits at this time while SQ programming guide has 3 bits
-var SQ_WAVE_LDS_ALLOC_VGPR_SHARED_SIZE_SHIFT    = 24
-var SQ_WAVE_LDS_ALLOC_VGPR_SHARED_SIZE_SIZE     = 4
-var SQ_WAVE_IB_STS2_WAVE64_SHIFT        = 11
-var SQ_WAVE_IB_STS2_WAVE64_SIZE         = 1
-
-var	SQ_WAVE_TRAPSTS_SAVECTX_MASK	=	0x400
-var SQ_WAVE_TRAPSTS_EXCE_MASK       =   0x1FF          			// Exception mask
-var	SQ_WAVE_TRAPSTS_SAVECTX_SHIFT	=	10					
-var	SQ_WAVE_TRAPSTS_MEM_VIOL_MASK	=	0x100					
-var	SQ_WAVE_TRAPSTS_MEM_VIOL_SHIFT	=	8		
-var	SQ_WAVE_TRAPSTS_PRE_SAVECTX_MASK 	=	0x3FF
-var	SQ_WAVE_TRAPSTS_PRE_SAVECTX_SHIFT 	=	0x0
-var	SQ_WAVE_TRAPSTS_PRE_SAVECTX_SIZE 	=	10
-var	SQ_WAVE_TRAPSTS_POST_SAVECTX_MASK 	=	0xFFFFF800	
-var	SQ_WAVE_TRAPSTS_POST_SAVECTX_SHIFT 	=	11
-var	SQ_WAVE_TRAPSTS_POST_SAVECTX_SIZE 	=	21	
-
-var SQ_WAVE_IB_STS_RCNT_SHIFT			=	16					//FIXME
-var SQ_WAVE_IB_STS_FIRST_REPLAY_SHIFT	=	15					//FIXME
-var SQ_WAVE_IB_STS_FIRST_REPLAY_SIZE    =   1                   //FIXME
-var SQ_WAVE_IB_STS_RCNT_SIZE            =   6                   //FIXME
-var SQ_WAVE_IB_STS_RCNT_FIRST_REPLAY_MASK_NEG	= 0x00007FFF	//FIXME
- 
-var	SQ_BUF_RSRC_WORD1_ATC_SHIFT		=	24
-var	SQ_BUF_RSRC_WORD3_MTYPE_SHIFT	=	27
-
-
-/*      Save        */
-var	S_SAVE_BUF_RSRC_WORD1_STRIDE		=	0x00040000  		//stride is 4 bytes 
-var	S_SAVE_BUF_RSRC_WORD3_MISC			= 	0x00807FAC			//SQ_SEL_X/Y/Z/W, BUF_NUM_FORMAT_FLOAT, (0 for MUBUF stride[17:14] when ADD_TID_ENABLE and BUF_DATA_FORMAT_32 for MTBUF), ADD_TID_ENABLE			
-
-var	S_SAVE_SPI_INIT_ATC_MASK			=	0x08000000			//bit[27]: ATC bit
-var	S_SAVE_SPI_INIT_ATC_SHIFT			=	27
-var	S_SAVE_SPI_INIT_MTYPE_MASK			=	0x70000000			//bit[30:28]: Mtype
-var	S_SAVE_SPI_INIT_MTYPE_SHIFT			=	28
-var	S_SAVE_SPI_INIT_FIRST_WAVE_MASK		=	0x04000000			//bit[26]: FirstWaveInTG
-var	S_SAVE_SPI_INIT_FIRST_WAVE_SHIFT	=	26
-
-var S_SAVE_PC_HI_RCNT_SHIFT				=	28					//FIXME	 check with Brian to ensure all fields other than PC[47:0] can be used
-var S_SAVE_PC_HI_RCNT_MASK				=   0xF0000000			//FIXME
-var S_SAVE_PC_HI_FIRST_REPLAY_SHIFT		=	27					//FIXME
-var S_SAVE_PC_HI_FIRST_REPLAY_MASK		=	0x08000000			//FIXME
-
-var	s_save_spi_init_lo				=	exec_lo
-var s_save_spi_init_hi				=	exec_hi
-
-var	s_save_pc_lo			=	ttmp0			//{TTMP1, TTMP0} = {3¡¯h0,pc_rewind[3:0], HT[0],trapID[7:0], PC[47:0]}
-var	s_save_pc_hi			=	ttmp1			
-var s_save_exec_lo			=	ttmp2
-var s_save_exec_hi			= 	ttmp3			
-var	s_save_status			=	ttmp4			
-var	s_save_trapsts			=	ttmp5			//not really used until the end of the SAVE routine
-var s_wave_size         	=	ttmp6           //ttmp6 is not needed now, since it's only 32bit xnack mask, now use it to determine wave32 or wave64 in EMU_HACK
-var s_save_xnack_mask	    =	ttmp7
-var	s_save_buf_rsrc0		=	ttmp8
-var	s_save_buf_rsrc1		=	ttmp9
-var	s_save_buf_rsrc2		=	ttmp10
-var	s_save_buf_rsrc3		=	ttmp11
-
-var s_save_mem_offset		= 	ttmp14
-var s_sgpr_save_num         =   106                     //in gfx10, all sgpr must be saved
-var s_save_alloc_size		=	s_save_trapsts			//conflict
-var s_save_tmp              =   s_save_buf_rsrc2       	//shared with s_save_buf_rsrc2  (conflict: should not use mem access with s_save_tmp at the same time)
-var s_save_m0				=	ttmp15					
-
-/*      Restore     */
-var	S_RESTORE_BUF_RSRC_WORD1_STRIDE			=	S_SAVE_BUF_RSRC_WORD1_STRIDE 
-var	S_RESTORE_BUF_RSRC_WORD3_MISC			= 	S_SAVE_BUF_RSRC_WORD3_MISC		 
-
-var	S_RESTORE_SPI_INIT_ATC_MASK			    =	0x08000000			//bit[27]: ATC bit
-var	S_RESTORE_SPI_INIT_ATC_SHIFT			=	27
-var	S_RESTORE_SPI_INIT_MTYPE_MASK			=	0x70000000			//bit[30:28]: Mtype
-var	S_RESTORE_SPI_INIT_MTYPE_SHIFT			=	28
-var	S_RESTORE_SPI_INIT_FIRST_WAVE_MASK		=	0x04000000			//bit[26]: FirstWaveInTG
-var	S_RESTORE_SPI_INIT_FIRST_WAVE_SHIFT	    =	26
-
-var S_RESTORE_PC_HI_RCNT_SHIFT				=	S_SAVE_PC_HI_RCNT_SHIFT
-var S_RESTORE_PC_HI_RCNT_MASK				=   S_SAVE_PC_HI_RCNT_MASK
-var S_RESTORE_PC_HI_FIRST_REPLAY_SHIFT		=	S_SAVE_PC_HI_FIRST_REPLAY_SHIFT
-var S_RESTORE_PC_HI_FIRST_REPLAY_MASK		=	S_SAVE_PC_HI_FIRST_REPLAY_MASK
-
-var s_restore_spi_init_lo                   =   exec_lo
-var s_restore_spi_init_hi                   =   exec_hi
-
-var s_restore_mem_offset		= 	ttmp12
-var s_restore_alloc_size		=	ttmp3
-var s_restore_tmp           	=   ttmp6
-var s_restore_mem_offset_save	= 	s_restore_tmp 		//no conflict
-
-var s_restore_m0			=	s_restore_alloc_size	//no conflict			
-
-var s_restore_mode			=  	ttmp13
-var s_restore_hwid1         =  ttmp2
-var s_restore_ddid          =  s_restore_hwid1
-var	s_restore_pc_lo		    =	ttmp0			
-var	s_restore_pc_hi		    =	ttmp1
-var s_restore_exec_lo		=	ttmp14
-var s_restore_exec_hi		= 	ttmp15
-var	s_restore_status	    =	ttmp4			
-var	s_restore_trapsts	    =	ttmp5
-//var s_restore_xnack_mask_lo	=	xnack_mask_lo
-//var s_restore_xnack_mask_hi	=	xnack_mask_hi
-var s_restore_xnack_mask    =   ttmp7
-var	s_restore_buf_rsrc0		=	ttmp8
-var	s_restore_buf_rsrc1		=	ttmp9
-var	s_restore_buf_rsrc2		=	ttmp10
-var	s_restore_buf_rsrc3		=	ttmp11
-var s_restore_size         	=	ttmp13                  //ttmp13 has no conflict
-
-/**************************************************************************/
-/*                     	trap handler entry points			              */
-/**************************************************************************/
-    if ((EMU_RUN_HACK) && (!EMU_RUN_HACK_RESTORE_NORMAL)) 					//hack to use trap_id for determining save/restore
-		//FIXME VCCZ un-init assertion s_getreg_b32  	s_save_status, hwreg(HW_REG_STATUS)			//save STATUS since we will change SCC
-		s_and_b32 s_save_tmp, s_save_pc_hi, 0xffff0000 				//change SCC
-    	s_cmp_eq_u32 s_save_tmp, 0x007e0000  						//Save: trap_id = 0x7e. Restore: trap_id = 0x7f.  
-    	s_cbranch_scc0 L_JUMP_TO_RESTORE							//do not need to recover STATUS here  since we are going to RESTORE
-		//FIXME  s_setreg_b32 	hwreg(HW_REG_STATUS), 	s_save_status		//need to recover STATUS since we are going to SAVE	
-		s_branch L_SKIP_RESTORE 									//NOT restore, SAVE actually
-	else	
-		s_branch L_SKIP_RESTORE 									//NOT restore. might be a regular trap or save
-    end
+	s_branch	L_SKIP_RESTORE						//NOT restore. might be a regular trap or save
 
 L_JUMP_TO_RESTORE:
-    s_branch L_RESTORE												//restore
+	s_branch	L_RESTORE
 
 L_SKIP_RESTORE:
-	
-	s_getreg_b32  	s_save_status, hwreg(HW_REG_STATUS)								//save STATUS since we will change SCC
-    s_andn2_b32		s_save_status, s_save_status, SQ_WAVE_STATUS_SPI_PRIO_MASK      //check whether this is for save
-	s_getreg_b32  	s_save_trapsts, hwreg(HW_REG_TRAPSTS)    		 				
-	s_and_b32		s_save_trapsts, s_save_trapsts, SQ_WAVE_TRAPSTS_SAVECTX_MASK	//check whether this is for save  
-	s_cbranch_scc1	L_SAVE															//this is the operation for save
-
-    // *********    Handle non-CWSR traps       *******************
-    if (!EMU_RUN_HACK)
-		s_getreg_b32     s_save_trapsts, hwreg(HW_REG_TRAPSTS)
-		s_and_b32        s_save_trapsts, s_save_trapsts, SQ_WAVE_TRAPSTS_EXCE_MASK // Check whether it is an exception
-		s_cbranch_scc1  L_EXCP_CASE   // Exception, jump back to the shader program directly.
-		s_add_u32    ttmp0, ttmp0, 4   // S_TRAP case, add 4 to ttmp0 
-		
-		L_EXCP_CASE:
-		s_and_b32    ttmp1, ttmp1, 0xFFFF
-		s_rfe_b64    [ttmp0, ttmp1]
-	end
-    // *********        End handling of non-CWSR traps   *******************
-
-/**************************************************************************/
-/*                     	save routine						              */
-/**************************************************************************/
-
-L_SAVE:	
-	
+	s_getreg_b32	s_save_status, hwreg(HW_REG_STATUS)			//save STATUS since we will change SCC
+	s_andn2_b32	s_save_status, s_save_status, SQ_WAVE_STATUS_SPI_PRIO_MASK
+	s_getreg_b32	s_save_trapsts, hwreg(HW_REG_TRAPSTS)
+	s_and_b32	ttmp2, s_save_trapsts, SQ_WAVE_TRAPSTS_SAVECTX_MASK	//check whether this is for save
+	s_cbranch_scc1	L_SAVE
+
+	// If STATUS.MEM_VIOL is asserted then halt the wave to prevent
+	// the exception raising again and blocking context save.
+	s_and_b32	ttmp2, s_save_trapsts, SQ_WAVE_TRAPSTS_MEM_VIOL_MASK
+	s_cbranch_scc0	L_FETCH_2ND_TRAP
+	s_or_b32	s_save_status, s_save_status, SQ_WAVE_STATUS_HALT_MASK
+
+L_FETCH_2ND_TRAP:
+	// Preserve and clear scalar XNACK state before issuing scalar loads.
+	// Save IB_STS.REPLAY_W64H[25], RCNT[21:16], FIRST_REPLAY[15] into
+	// unused space ttmp11[31:24].
+	s_andn2_b32	ttmp11, ttmp11, (TTMP11_SAVE_REPLAY_W64H_MASK | TTMP11_SAVE_RCNT_FIRST_REPLAY_MASK)
+	s_getreg_b32	ttmp2, hwreg(HW_REG_IB_STS)
+	s_and_b32	ttmp3, ttmp2, SQ_WAVE_IB_STS_REPLAY_W64H_MASK
+	s_lshl_b32	ttmp3, ttmp3, (TTMP11_SAVE_REPLAY_W64H_SHIFT - SQ_WAVE_IB_STS_REPLAY_W64H_SHIFT)
+	s_or_b32	ttmp11, ttmp11, ttmp3
+	s_and_b32	ttmp3, ttmp2, SQ_WAVE_IB_STS_RCNT_FIRST_REPLAY_MASK
+	s_lshl_b32	ttmp3, ttmp3, (TTMP11_SAVE_RCNT_FIRST_REPLAY_SHIFT - SQ_WAVE_IB_STS_FIRST_REPLAY_SHIFT)
+	s_or_b32	ttmp11, ttmp11, ttmp3
+	s_andn2_b32	ttmp2, ttmp2, (SQ_WAVE_IB_STS_REPLAY_W64H_MASK | SQ_WAVE_IB_STS_RCNT_FIRST_REPLAY_MASK)
+	s_setreg_b32	hwreg(HW_REG_IB_STS), ttmp2
+
+	// Read second-level TBA/TMA from first-level TMA and jump if available.
+	// ttmp[2:5] and ttmp12 can be used (others hold SPI-initialized debug data)
+	// ttmp12 holds SQ_WAVE_STATUS
+	s_getreg_b32	ttmp4, hwreg(HW_REG_SHADER_TMA_LO)
+	s_getreg_b32	ttmp5, hwreg(HW_REG_SHADER_TMA_HI)
+	s_lshl_b64	[ttmp4, ttmp5], [ttmp4, ttmp5], 0x8
+	s_load_dwordx2	[ttmp2, ttmp3], [ttmp4, ttmp5], 0x0 glc:1		// second-level TBA
+	s_waitcnt	lgkmcnt(0)
+	s_load_dwordx2	[ttmp4, ttmp5], [ttmp4, ttmp5], 0x8 glc:1		// second-level TMA
+	s_waitcnt	lgkmcnt(0)
+	s_and_b64	[ttmp2, ttmp3], [ttmp2, ttmp3], [ttmp2, ttmp3]
+	s_cbranch_scc0	L_NO_NEXT_TRAP						// second-level trap handler not been set
+	s_setpc_b64	[ttmp2, ttmp3]						// jump to second-level trap handler
+
+L_NO_NEXT_TRAP:
+	s_getreg_b32	s_save_trapsts, hwreg(HW_REG_TRAPSTS)
+	s_and_b32	s_save_trapsts, s_save_trapsts, SQ_WAVE_TRAPSTS_EXCE_MASK
+	s_cbranch_scc1	L_EXCP_CASE						// Exception, jump back to the shader program directly.
+	s_add_u32	ttmp0, ttmp0, 4						// S_TRAP case, add 4 to ttmp0
+	s_addc_u32	ttmp1, ttmp1, 0
+L_EXCP_CASE:
+	s_and_b32	ttmp1, ttmp1, 0xFFFF
+
+	// Restore SQ_WAVE_IB_STS.
+	s_lshr_b32	ttmp2, ttmp11, (TTMP11_SAVE_RCNT_FIRST_REPLAY_SHIFT - SQ_WAVE_IB_STS_FIRST_REPLAY_SHIFT)
+	s_and_b32	ttmp3, ttmp2, SQ_WAVE_IB_STS_RCNT_FIRST_REPLAY_MASK
+	s_lshr_b32	ttmp2, ttmp11, (TTMP11_SAVE_REPLAY_W64H_SHIFT - SQ_WAVE_IB_STS_REPLAY_W64H_SHIFT)
+	s_and_b32	ttmp2, ttmp2, SQ_WAVE_IB_STS_REPLAY_W64H_MASK
+	s_or_b32	ttmp2, ttmp2, ttmp3
+	s_setreg_b32	hwreg(HW_REG_IB_STS), ttmp2
+
+	// Restore SQ_WAVE_STATUS.
+	s_and_b64	exec, exec, exec					// Restore STATUS.EXECZ, not writable by s_setreg_b32
+	s_and_b64	vcc, vcc, vcc						// Restore STATUS.VCCZ, not writable by s_setreg_b32
+	s_setreg_b32	hwreg(HW_REG_STATUS), s_save_status
+
+	s_rfe_b64	[ttmp0, ttmp1]
+
+L_SAVE:
 	//check whether there is mem_viol
-	s_getreg_b32  	s_save_trapsts, hwreg(HW_REG_TRAPSTS)    		 				
-	s_and_b32		s_save_trapsts, s_save_trapsts, SQ_WAVE_TRAPSTS_MEM_VIOL_MASK			
+	s_getreg_b32	s_save_trapsts, hwreg(HW_REG_TRAPSTS)
+	s_and_b32	s_save_trapsts, s_save_trapsts, SQ_WAVE_TRAPSTS_MEM_VIOL_MASK
 	s_cbranch_scc0	L_NO_PC_REWIND
-    
+
 	//if so, need rewind PC assuming GDS operation gets NACKed
-	s_mov_b32       s_save_tmp, 0															//clear mem_viol bit
-	s_setreg_b32	hwreg(HW_REG_TRAPSTS, SQ_WAVE_TRAPSTS_MEM_VIOL_SHIFT, 1), s_save_tmp	//clear mem_viol bit 
-	s_and_b32 		s_save_pc_hi, s_save_pc_hi, 0x0000ffff    //pc[47:32]
-	s_sub_u32 		s_save_pc_lo, s_save_pc_lo, 8             //pc[31:0]-8
-	s_subb_u32 		s_save_pc_hi, s_save_pc_hi, 0x0			  // -scc
+	s_mov_b32	s_save_tmp, 0
+	s_setreg_b32	hwreg(HW_REG_TRAPSTS, SQ_WAVE_TRAPSTS_MEM_VIOL_SHIFT, 1), s_save_tmp	//clear mem_viol bit
+	s_and_b32	s_save_pc_hi, s_save_pc_hi, 0x0000ffff			//pc[47:32]
+	s_sub_u32	s_save_pc_lo, s_save_pc_lo, 8				//pc[31:0]-8
+	s_subb_u32	s_save_pc_hi, s_save_pc_hi, 0x0
 
 L_NO_PC_REWIND:
-    s_mov_b32       s_save_tmp, 0															//clear saveCtx bit
-	s_setreg_b32	hwreg(HW_REG_TRAPSTS, SQ_WAVE_TRAPSTS_SAVECTX_SHIFT, 1), s_save_tmp		//clear saveCtx bit   
-
-	//s_mov_b32		s_save_xnack_mask_lo,	xnack_mask_lo									//save XNACK_MASK  
-	//s_mov_b32		s_save_xnack_mask_hi,	xnack_mask_hi
-    s_getreg_b32	s_save_xnack_mask,  hwreg(HW_REG_SHADER_XNACK_MASK)  
-	s_getreg_b32	s_save_tmp, hwreg(HW_REG_IB_STS, SQ_WAVE_IB_STS_RCNT_SHIFT, SQ_WAVE_IB_STS_RCNT_SIZE)					//save RCNT
-	s_lshl_b32		s_save_tmp, s_save_tmp, S_SAVE_PC_HI_RCNT_SHIFT
-	s_or_b32		s_save_pc_hi, s_save_pc_hi, s_save_tmp
-	s_getreg_b32	s_save_tmp, hwreg(HW_REG_IB_STS, SQ_WAVE_IB_STS_FIRST_REPLAY_SHIFT, SQ_WAVE_IB_STS_FIRST_REPLAY_SIZE)	//save FIRST_REPLAY
-	s_lshl_b32		s_save_tmp, s_save_tmp, S_SAVE_PC_HI_FIRST_REPLAY_SHIFT
-	s_or_b32		s_save_pc_hi, s_save_pc_hi, s_save_tmp
-	s_getreg_b32	s_save_tmp, hwreg(HW_REG_IB_STS)										//clear RCNT and FIRST_REPLAY in IB_STS
-	s_and_b32		s_save_tmp, s_save_tmp, SQ_WAVE_IB_STS_RCNT_FIRST_REPLAY_MASK_NEG
+	s_mov_b32	s_save_tmp, 0
+	s_setreg_b32	hwreg(HW_REG_TRAPSTS, SQ_WAVE_TRAPSTS_SAVECTX_SHIFT, 1), s_save_tmp	//clear saveCtx bit
+
+	s_getreg_b32	s_save_xnack_mask, hwreg(HW_REG_SHADER_XNACK_MASK)
+	s_getreg_b32	s_save_tmp, hwreg(HW_REG_IB_STS, SQ_WAVE_IB_STS_RCNT_SHIFT, SQ_WAVE_IB_STS_RCNT_SIZE)
+	s_lshl_b32	s_save_tmp, s_save_tmp, S_SAVE_PC_HI_RCNT_SHIFT
+	s_or_b32	s_save_pc_hi, s_save_pc_hi, s_save_tmp
+	s_getreg_b32	s_save_tmp, hwreg(HW_REG_IB_STS, SQ_WAVE_IB_STS_FIRST_REPLAY_SHIFT, SQ_WAVE_IB_STS_FIRST_REPLAY_SIZE)
+	s_lshl_b32	s_save_tmp, s_save_tmp, S_SAVE_PC_HI_FIRST_REPLAY_SHIFT
+	s_or_b32	s_save_pc_hi, s_save_pc_hi, s_save_tmp
+	s_getreg_b32	s_save_tmp, hwreg(HW_REG_IB_STS, SQ_WAVE_IB_STS_REPLAY_W64H_SHIFT, SQ_WAVE_IB_STS_REPLAY_W64H_SIZE)
+	s_lshl_b32	s_save_tmp, s_save_tmp, S_SAVE_PC_HI_REPLAY_W64H_SHIFT
+	s_or_b32	s_save_pc_hi, s_save_pc_hi, s_save_tmp
+	s_getreg_b32	s_save_tmp, hwreg(HW_REG_IB_STS)			//clear RCNT and FIRST_REPLAY and REPLAY_W64H in IB_STS
+	s_and_b32	s_save_tmp, s_save_tmp, SQ_WAVE_IB_STS_RCNT_FIRST_REPLAY_MASK_NEG
 
 	s_setreg_b32	hwreg(HW_REG_IB_STS), s_save_tmp
-    
-	/*		inform SPI the readiness and wait for SPI's go signal */
-	s_mov_b32		s_save_exec_lo,	exec_lo													//save EXEC and use EXEC for the go signal from SPI
-	s_mov_b32		s_save_exec_hi,	exec_hi
-	s_mov_b64		exec, 	0x0																//clear EXEC to get ready to receive
-	if (EMU_RUN_HACK)
-	
-	else
-		s_sendmsg	sendmsg(MSG_SAVEWAVE)													//send SPI a message and wait for SPI's write to EXEC  
-	end
-
-  L_SLEEP:		
-	s_sleep 0x2
-	
-	if (EMU_RUN_HACK)
-																							
-	else
-		s_cbranch_execz	L_SLEEP                                                         
-	end
-
-
-	/*      setup Resource Contants    */
-	if ((EMU_RUN_HACK) && (!EMU_RUN_HACK_SAVE_SINGLE_WAVE))	
-		//calculate wd_addr using absolute thread id 
-		v_readlane_b32 s_save_tmp, v9, 0
-        //determine it is wave32 or wave64
-        s_getreg_b32 	s_wave_size, hwreg(HW_REG_IB_STS2,SQ_WAVE_IB_STS2_WAVE64_SHIFT,SQ_WAVE_IB_STS2_WAVE64_SIZE)
-        s_cmp_eq_u32    s_wave_size, 0
-        s_cbranch_scc1  L_SAVE_WAVE32
-        s_lshr_b32 s_save_tmp, s_save_tmp, 6 //SAVE WAVE64
-        s_branch    L_SAVE_CON
-    L_SAVE_WAVE32:
-        s_lshr_b32 s_save_tmp, s_save_tmp, 5 //SAVE WAVE32
-    L_SAVE_CON:
-		s_mul_i32 s_save_tmp, s_save_tmp, WAVE_SPACE
-		s_add_i32 s_save_spi_init_lo, s_save_tmp, WG_BASE_ADDR_LO
-		s_mov_b32 s_save_spi_init_hi, WG_BASE_ADDR_HI
-		s_and_b32 s_save_spi_init_hi, s_save_spi_init_hi, CTX_SAVE_CONTROL		
-	else
-	end
-	if ((EMU_RUN_HACK) && (EMU_RUN_HACK_SAVE_SINGLE_WAVE))
-		s_add_i32 s_save_spi_init_lo, s_save_tmp, WG_BASE_ADDR_LO
-		s_mov_b32 s_save_spi_init_hi, WG_BASE_ADDR_HI
-		s_and_b32 s_save_spi_init_hi, s_save_spi_init_hi, CTX_SAVE_CONTROL		
-	else
-	end
-	
-	
-	s_mov_b32		s_save_buf_rsrc0, 	s_save_spi_init_lo														//base_addr_lo
-	s_and_b32		s_save_buf_rsrc1, 	s_save_spi_init_hi, 0x0000FFFF											//base_addr_hi
-	s_or_b32		s_save_buf_rsrc1, 	s_save_buf_rsrc1,  S_SAVE_BUF_RSRC_WORD1_STRIDE
-    s_mov_b32       s_save_buf_rsrc2,   0                                               						//NUM_RECORDS initial value = 0 (in bytes) although not neccessarily inited
-	s_mov_b32		s_save_buf_rsrc3, 	S_SAVE_BUF_RSRC_WORD3_MISC
-	s_and_b32		s_save_tmp,         s_save_spi_init_hi, S_SAVE_SPI_INIT_ATC_MASK		
-	s_lshr_b32		s_save_tmp,  		s_save_tmp, (S_SAVE_SPI_INIT_ATC_SHIFT-SQ_BUF_RSRC_WORD1_ATC_SHIFT)			//get ATC bit into position
-	s_or_b32		s_save_buf_rsrc3, 	s_save_buf_rsrc3,  s_save_tmp											//or ATC
-	s_and_b32		s_save_tmp,         s_save_spi_init_hi, S_SAVE_SPI_INIT_MTYPE_MASK		
-	s_lshr_b32		s_save_tmp,  		s_save_tmp, (S_SAVE_SPI_INIT_MTYPE_SHIFT-SQ_BUF_RSRC_WORD3_MTYPE_SHIFT)		//get MTYPE bits into position
-	s_or_b32		s_save_buf_rsrc3, 	s_save_buf_rsrc3,  s_save_tmp											//or MTYPE	
-	
-	s_mov_b32		s_save_m0,			m0																	//save M0
-	
-	/* 		global mem offset			*/
-	s_mov_b32		s_save_mem_offset, 	0x0																		//mem offset initial value = 0
-    s_getreg_b32 	s_wave_size, hwreg(HW_REG_IB_STS2,SQ_WAVE_IB_STS2_WAVE64_SHIFT,SQ_WAVE_IB_STS2_WAVE64_SIZE) //get wave_save_size
-    s_or_b32        s_wave_size, s_save_spi_init_hi,    s_wave_size                                             //share s_wave_size with exec_hi
-
-    /*      	save VGPRs	    */
-	//////////////////////////////
-  L_SAVE_VGPR:
-  
- 	s_mov_b32		exec_lo, 0xFFFFFFFF 											//need every thread from now on
-    s_and_b32       m0, s_wave_size, 1
-    s_cmp_eq_u32    m0, 1  
-    s_cbranch_scc1  L_ENABLE_SAVE_VGPR_EXEC_HI   
-    s_mov_b32		exec_hi, 0x00000000
-    s_branch        L_SAVE_VGPR_NORMAL
-  L_ENABLE_SAVE_VGPR_EXEC_HI:
-	s_mov_b32		exec_hi, 0xFFFFFFFF
-  L_SAVE_VGPR_NORMAL:	
-	s_getreg_b32 	s_save_alloc_size, hwreg(HW_REG_GPR_ALLOC,SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SHIFT,SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SIZE) 					//vpgr_size
-	//for wave32 and wave64, the num of vgpr function is the same?
-    s_add_u32 		s_save_alloc_size, s_save_alloc_size, 1
-	s_lshl_b32 		s_save_alloc_size, s_save_alloc_size, 2 						//Number of VGPRs = (vgpr_size + 1) * 4    (non-zero value)   //FIXME for GFX, zero is possible
-    //determine it is wave32 or wave64
-    s_and_b32       m0, s_wave_size, 1
-    s_cmp_eq_u32    m0, 1
-    s_cbranch_scc1  L_SAVE_VGPR_WAVE64
-
-    //zhenxu added it for save vgpr for wave32
-	s_lshl_b32		s_save_buf_rsrc2,  s_save_alloc_size, 7							//NUM_RECORDS in bytes (32 threads*4)
-	if (SWIZZLE_EN)
-		s_add_u32		s_save_buf_rsrc2, s_save_buf_rsrc2, 0x0						//FIXME need to use swizzle to enable bounds checking?
-	else
-		s_mov_b32		s_save_buf_rsrc2,  0x1000000								//NUM_RECORDS in bytes
-	end
-	
-    s_mov_b32 		m0, 0x0 														//VGPR initial index value =0
-	//s_set_gpr_idx_on  m0, 0x1														//M0[7:0] = M0[7:0] and M0[15:12] = 0x1
-    //s_add_u32		s_save_alloc_size, s_save_alloc_size, 0x1000					//add 0x1000 since we compare m0 against it later, doesn't need this in gfx10
-
-  L_SAVE_VGPR_WAVE32_LOOP: 										
-	v_movrels_b32 		v0, v0															//v0 = v[0+m0]	
-	    
-    if(USE_MTBUF_INSTEAD_OF_MUBUF)       
-		tbuffer_store_format_x v0, v0, s_save_buf_rsrc0, s_save_mem_offset format:BUF_NUM_FORMAT_FLOAT format: BUF_DATA_FORMAT_32 slc:1 glc:1
-    else
-		buffer_store_dword v0, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1
-	end
-
-    s_add_u32		m0, m0, 1														//next vgpr index
-	s_add_u32		s_save_mem_offset, s_save_mem_offset, 128						//every buffer_store_dword does 128 bytes
-	s_cmp_lt_u32 	m0,	s_save_alloc_size 											//scc = (m0 < s_save_alloc_size) ? 1 : 0
-	s_cbranch_scc1 	L_SAVE_VGPR_WAVE32_LOOP												//VGPR save is complete?
-    s_branch    L_SAVE_LDS
-    //save vgpr for wave32 ends
-
-  L_SAVE_VGPR_WAVE64:
-	s_lshl_b32		s_save_buf_rsrc2,  s_save_alloc_size, 8							//NUM_RECORDS in bytes (64 threads*4)
-	if (SWIZZLE_EN)
-		s_add_u32		s_save_buf_rsrc2, s_save_buf_rsrc2, 0x0						//FIXME need to use swizzle to enable bounds checking?
-	else
-		s_mov_b32		s_save_buf_rsrc2,  0x1000000								//NUM_RECORDS in bytes
-	end
-	
-    s_mov_b32 		m0, 0x0 														//VGPR initial index value =0
-	//s_set_gpr_idx_on  m0, 0x1														//M0[7:0] = M0[7:0] and M0[15:12] = 0x1
-    //s_add_u32		s_save_alloc_size, s_save_alloc_size, 0x1000					//add 0x1000 since we compare m0 against it later, doesn't need this in gfx10
-
-  L_SAVE_VGPR_WAVE64_LOOP: 										
-	v_movrels_b32 		v0, v0															//v0 = v[0+m0]	
-	    
-    if(USE_MTBUF_INSTEAD_OF_MUBUF)       
-		tbuffer_store_format_x v0, v0, s_save_buf_rsrc0, s_save_mem_offset format:BUF_NUM_FORMAT_FLOAT format: BUF_DATA_FORMAT_32 slc:1 glc:1
-    else
-		buffer_store_dword v0, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1
-	end
-
-    s_add_u32		m0, m0, 1														//next vgpr index
-	s_add_u32		s_save_mem_offset, s_save_mem_offset, 256						//every buffer_store_dword does 256 bytes
-	s_cmp_lt_u32 	m0,	s_save_alloc_size 											//scc = (m0 < s_save_alloc_size) ? 1 : 0
-	s_cbranch_scc1 	L_SAVE_VGPR_WAVE64_LOOP												//VGPR save is complete?
-	//s_set_gpr_idx_off
-    //
-    //Below part will be the save shared vgpr part (new for gfx10)
-    s_getreg_b32 	s_save_alloc_size, hwreg(HW_REG_LDS_ALLOC,SQ_WAVE_LDS_ALLOC_VGPR_SHARED_SIZE_SHIFT,SQ_WAVE_LDS_ALLOC_VGPR_SHARED_SIZE_SIZE) 			//shared_vgpr_size
-    s_and_b32		s_save_alloc_size, s_save_alloc_size, 0xFFFFFFFF				//shared_vgpr_size is zero?
-    s_cbranch_scc0	L_SAVE_LDS													    //no shared_vgpr used? jump to L_SAVE_LDS
-    s_lshl_b32 		s_save_alloc_size, s_save_alloc_size, 3 						//Number of SHARED_VGPRs = shared_vgpr_size * 8    (non-zero value)
-    //m0 now has the value of normal vgpr count, just add the m0 with shared_vgpr count to get the total count.
-    //save shared_vgpr will start from the index of m0
-    s_add_u32       s_save_alloc_size, s_save_alloc_size, m0
-    s_mov_b32		exec_lo, 0xFFFFFFFF
-    s_mov_b32		exec_hi, 0x00000000
-    L_SAVE_SHARED_VGPR_WAVE64_LOOP: 										
-	v_movrels_b32 		v0, v0															//v0 = v[0+m0]	
-	buffer_store_dword v0, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1
-    s_add_u32		m0, m0, 1														//next vgpr index
-	s_add_u32		s_save_mem_offset, s_save_mem_offset, 128						//every buffer_store_dword does 256 bytes
-	s_cmp_lt_u32 	m0,	s_save_alloc_size 											//scc = (m0 < s_save_alloc_size) ? 1 : 0
-	s_cbranch_scc1 	L_SAVE_SHARED_VGPR_WAVE64_LOOP									//SHARED_VGPR save is complete?
-    
-	/*      	save LDS	    */
-	//////////////////////////////
-  L_SAVE_LDS:
-
-    //Only check the first wave need LDS
-	/*      the first wave in the threadgroup    */
-	s_barrier																		//FIXME  not performance-optimal "LDS is used? wait for other waves in the same TG" 
-	s_and_b32		s_save_tmp, s_wave_size, S_SAVE_SPI_INIT_FIRST_WAVE_MASK								//exec is still used here
-	s_cbranch_scc0	L_SAVE_SGPR
-	
-	s_mov_b32		exec_lo, 0xFFFFFFFF 											//need every thread from now on
-    s_and_b32       m0, s_wave_size, 1
-    s_cmp_eq_u32    m0, 1
-    s_cbranch_scc1  L_ENABLE_SAVE_LDS_EXEC_HI   
-    s_mov_b32		exec_hi, 0x00000000
-    s_branch        L_SAVE_LDS_NORMAL
-  L_ENABLE_SAVE_LDS_EXEC_HI:
-	s_mov_b32		exec_hi, 0xFFFFFFFF
-  L_SAVE_LDS_NORMAL:	
-	s_getreg_b32 	s_save_alloc_size, hwreg(HW_REG_LDS_ALLOC,SQ_WAVE_LDS_ALLOC_LDS_SIZE_SHIFT,SQ_WAVE_LDS_ALLOC_LDS_SIZE_SIZE) 			//lds_size
-	s_and_b32		s_save_alloc_size, s_save_alloc_size, 0xFFFFFFFF				//lds_size is zero?
-	s_cbranch_scc0	L_SAVE_SGPR														//no lds used? jump to L_SAVE_VGPR
-	s_lshl_b32 		s_save_alloc_size, s_save_alloc_size, 6 						//LDS size in dwords = lds_size * 64dw
-	s_lshl_b32 		s_save_alloc_size, s_save_alloc_size, 2 						//LDS size in bytes
-	s_mov_b32		s_save_buf_rsrc2,  s_save_alloc_size  							//NUM_RECORDS in bytes
-	if (SWIZZLE_EN)
-		s_add_u32		s_save_buf_rsrc2, s_save_buf_rsrc2, 0x0						//FIXME need to use swizzle to enable bounds checking?
-	else
-		s_mov_b32		s_save_buf_rsrc2,  0x1000000								//NUM_RECORDS in bytes
-	end
-
-    //load 0~63*4(byte address) to vgpr v15
-    v_mbcnt_lo_u32_b32 v0, -1, 0
-    v_mbcnt_hi_u32_b32 v0, -1, v0
-    v_mul_u32_u24 v0, 4, v0
-
-    s_and_b32       m0, s_wave_size, 1
-    s_cmp_eq_u32    m0, 1
-    s_mov_b32 		m0, 0x0
-    s_cbranch_scc1  L_SAVE_LDS_LOOP_W64
-
-  L_SAVE_LDS_LOOP_W32:									
-	if (SAVE_LDS)
-    ds_read_b32 v1, v0
-    s_waitcnt 0														    //ensure data ready
-    buffer_store_dword v1, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1
-	//buffer_store_lds_dword	s_save_buf_rsrc0, s_save_mem_offset lds:1               //save lds to memory doesn't exist in 10
-	end
-	s_add_u32		m0, m0, 128															//every buffer_store_lds does 128 bytes
-	s_add_u32		s_save_mem_offset, s_save_mem_offset, 128							//mem offset increased by 128 bytes
-    v_add_nc_u32    v0, v0, 128
-	s_cmp_lt_u32	m0, s_save_alloc_size												//scc=(m0 < s_save_alloc_size) ? 1 : 0
-	s_cbranch_scc1  L_SAVE_LDS_LOOP_W32													//LDS save is complete?
-    s_branch        L_SAVE_SGPR
-
-  L_SAVE_LDS_LOOP_W64:									
-	if (SAVE_LDS)
-    ds_read_b32 v1, v0
-    s_waitcnt 0														    //ensure data ready
-    buffer_store_dword v1, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1
-	//buffer_store_lds_dword	s_save_buf_rsrc0, s_save_mem_offset lds:1               //save lds to memory doesn't exist in 10
-	end
-	s_add_u32		m0, m0, 256															//every buffer_store_lds does 256 bytes
-	s_add_u32		s_save_mem_offset, s_save_mem_offset, 256							//mem offset increased by 256 bytes
-    v_add_nc_u32    v0, v0, 256
-	s_cmp_lt_u32	m0, s_save_alloc_size												//scc=(m0 < s_save_alloc_size) ? 1 : 0
-	s_cbranch_scc1  L_SAVE_LDS_LOOP_W64													//LDS save is complete?
-   
-	
-	/*      	save SGPRs	    */
-	//////////////////////////////
-	//s_getreg_b32 	s_save_alloc_size, hwreg(HW_REG_GPR_ALLOC,SQ_WAVE_GPR_ALLOC_SGPR_SIZE_SHIFT,SQ_WAVE_GPR_ALLOC_SGPR_SIZE_SIZE) 				//spgr_size
-	//s_add_u32 		s_save_alloc_size, s_save_alloc_size, 1
-	//s_lshl_b32 		s_save_alloc_size, s_save_alloc_size, 4 						//Number of SGPRs = (sgpr_size + 1) * 16   (non-zero value) 
-	//s_lshl_b32 		s_save_alloc_size, s_save_alloc_size, 3 						//In gfx10, Number of SGPRs = (sgpr_size + 1) * 8   (non-zero value) 
-  L_SAVE_SGPR:
-    //need to look at it is wave32 or wave64
-    s_and_b32       m0, s_wave_size, 1
-    s_cmp_eq_u32    m0, 1
-    s_cbranch_scc1  L_SAVE_SGPR_VMEM_WAVE64
-    if (SGPR_SAVE_USE_SQC)
-		s_lshl_b32		s_save_buf_rsrc2,	s_sgpr_save_num, 2					//NUM_RECORDS in bytes
-    else
-        s_lshl_b32		s_save_buf_rsrc2,	s_sgpr_save_num, 7					//NUM_RECORDS in bytes (32 threads)
-    end
-    s_branch    L_SAVE_SGPR_CONT    
-  L_SAVE_SGPR_VMEM_WAVE64:
-	if (SGPR_SAVE_USE_SQC)
-		s_lshl_b32		s_save_buf_rsrc2,	s_sgpr_save_num, 2					//NUM_RECORDS in bytes 
-	else
-		s_lshl_b32		s_save_buf_rsrc2,	s_sgpr_save_num, 8					//NUM_RECORDS in bytes (64 threads)
-	end
-  L_SAVE_SGPR_CONT:
-	if (SWIZZLE_EN)
-		s_add_u32		s_save_buf_rsrc2, s_save_buf_rsrc2, 0x0						//FIXME need to use swizzle to enable bounds checking?
-	else
-		s_mov_b32		s_save_buf_rsrc2,  0x1000000								//NUM_RECORDS in bytes
-	end
-	
-	//s_mov_b32 		m0, 0x0 														//SGPR initial index value =0		
-    //s_nop           0x0                                                             //Manually inserted wait states
-	
-    s_and_b32       m0, s_wave_size, 1
-    s_cmp_eq_u32    m0, 1
-    
-    s_mov_b32 		m0, 0x0 														//SGPR initial index value =0		
-    s_nop           0x0                                                             //Manually inserted wait states
-
-    s_cbranch_scc1  L_SAVE_SGPR_LOOP_WAVE64
-
-  L_SAVE_SGPR_LOOP_WAVE32: 										
-	s_movrels_b32 	s0, s0 															//s0 = s[0+m0]
-    //zhenxu, adding one more argument to save sgpr function, this is only for vmem, using sqc is not change    
-	write_sgpr_to_mem_wave32(s0, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)							//PV: the best performance should be using s_buffer_store_dwordx4
-	s_add_u32		m0, m0, 1														//next sgpr index
-	s_cmp_lt_u32 	m0, s_sgpr_save_num 											//scc = (m0 < s_sgpr_save_num) ? 1 : 0
-	s_cbranch_scc1 	L_SAVE_SGPR_LOOP_WAVE32												//SGPR save is complete?
-    s_branch    L_SAVE_HWREG
-
-  L_SAVE_SGPR_LOOP_WAVE64: 										
-	s_movrels_b32 	s0, s0 															//s0 = s[0+m0]
-    //zhenxu, adding one more argument to save sgpr function, this is only for vmem, using sqc is not change    
-	write_sgpr_to_mem_wave64(s0, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)							//PV: the best performance should be using s_buffer_store_dwordx4
-	s_add_u32		m0, m0, 1														//next sgpr index
-	s_cmp_lt_u32 	m0, s_sgpr_save_num 											//scc = (m0 < s_sgpr_save_num) ? 1 : 0
-	s_cbranch_scc1 	L_SAVE_SGPR_LOOP_WAVE64												//SGPR save is complete?
-
-	
-	/* 		save HW registers	*/
-	//////////////////////////////
-  L_SAVE_HWREG:
-    s_mov_b32		s_save_buf_rsrc2, 0x4								//NUM_RECORDS	in bytes
-	if (SWIZZLE_EN)
-		s_add_u32		s_save_buf_rsrc2, s_save_buf_rsrc2, 0x0						//FIXME need to use swizzle to enable bounds checking?
-	else
-		s_mov_b32		s_save_buf_rsrc2,  0x1000000								//NUM_RECORDS in bytes
-	end
-
-    s_and_b32       m0, s_wave_size, 1
-    s_cmp_eq_u32    m0, 1
-    s_cbranch_scc1  L_SAVE_HWREG_WAVE64
-	
-	write_sgpr_to_mem_wave32(s_save_m0, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)					//M0
-
-	if ((EMU_RUN_HACK) && (EMU_RUN_HACK_SAVE_FIRST_TIME))      
-		s_add_u32 s_save_pc_lo, s_save_pc_lo, 4             //pc[31:0]+4
-		s_addc_u32 s_save_pc_hi, s_save_pc_hi, 0x0			//carry bit over
-	end
-
-	write_sgpr_to_mem_wave32(s_save_pc_lo, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)					//PC
-	write_sgpr_to_mem_wave32(s_save_pc_hi, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)
-	write_sgpr_to_mem_wave32(s_save_exec_lo, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)				//EXEC
-	write_sgpr_to_mem_wave32(s_save_exec_hi, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)
-	write_sgpr_to_mem_wave32(s_save_status, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)				//STATUS 
-	
-	//s_save_trapsts conflicts with s_save_alloc_size
-	s_getreg_b32    s_save_trapsts, hwreg(HW_REG_TRAPSTS)
-	write_sgpr_to_mem_wave32(s_save_trapsts, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)				//TRAPSTS
-	
-	//write_sgpr_to_mem_wave32(s_save_xnack_mask_lo, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)			//XNACK_MASK_LO
-	write_sgpr_to_mem_wave32(s_save_xnack_mask, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)			//XNACK_MASK_HI
-	
-	//use s_save_tmp would introduce conflict here between s_save_tmp and s_save_buf_rsrc2
-	s_getreg_b32 	s_save_m0, hwreg(HW_REG_MODE)																						//MODE
-	write_sgpr_to_mem_wave32(s_save_m0, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)
-    if(SAVE_RESTORE_HWID_DDID)
-    s_getreg_b32 	s_save_m0, hwreg(HW_REG_HW_ID1)																						//HW_ID1, handler records the SE/SA/WGP/SIMD/wave of the original wave
-    write_sgpr_to_mem_wave32(s_save_m0, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)
-    end
-    s_branch   L_S_PGM_END_SAVED
-
-  L_SAVE_HWREG_WAVE64:
-    write_sgpr_to_mem_wave64(s_save_m0, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)					//M0
-
-	if ((EMU_RUN_HACK) && (EMU_RUN_HACK_SAVE_FIRST_TIME))      
-		s_add_u32 s_save_pc_lo, s_save_pc_lo, 4             //pc[31:0]+4
-		s_addc_u32 s_save_pc_hi, s_save_pc_hi, 0x0			//carry bit over
-	end
-
-	write_sgpr_to_mem_wave64(s_save_pc_lo, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)					//PC
-	write_sgpr_to_mem_wave64(s_save_pc_hi, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)
-	write_sgpr_to_mem_wave64(s_save_exec_lo, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)				//EXEC
-	write_sgpr_to_mem_wave64(s_save_exec_hi, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)
-	write_sgpr_to_mem_wave64(s_save_status, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)				//STATUS 
-	
-	//s_save_trapsts conflicts with s_save_alloc_size
-	s_getreg_b32    s_save_trapsts, hwreg(HW_REG_TRAPSTS)
-	write_sgpr_to_mem_wave64(s_save_trapsts, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)				//TRAPSTS
-	
-	//write_sgpr_to_mem_wave64(s_save_xnack_mask_lo, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)			//XNACK_MASK_LO
-	write_sgpr_to_mem_wave64(s_save_xnack_mask, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)			//XNACK_MASK_HI
-	
-	//use s_save_tmp would introduce conflict here between s_save_tmp and s_save_buf_rsrc2
-	s_getreg_b32 	s_save_m0, hwreg(HW_REG_MODE)																						//MODE
-	write_sgpr_to_mem_wave64(s_save_m0, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)
-
-
-    if(SAVE_RESTORE_HWID_DDID)
-    s_getreg_b32 	s_save_m0, hwreg(HW_REG_HW_ID1)																						//HW_ID1, handler records the SE/SA/WGP/SIMD/wave of the original wave
-    write_sgpr_to_mem_wave64(s_save_m0, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF)
-
-	/* 		save DDID	*/
-	//////////////////////////////
-  L_SAVE_DDID:
-    //EXEC has been saved, no vector inst following
-    s_mov_b32	exec_lo, 0x80000000    //Set MSB to 1. Cleared when draw index is returned
-    s_sendmsg sendmsg(MSG_GET_DDID)
-
-  L_WAIT_DDID_LOOP:    
-    s_nop		7			// sleep a bit
-    s_bitcmp0_b32 exec_lo, 31	// test to see if MSB is cleared, meaning done
-    s_cbranch_scc0	L_WAIT_DDID_LOOP
-
-    s_mov_b32	s_save_m0, exec_lo
-
-
-    s_mov_b32		s_save_buf_rsrc2, 0x4								//NUM_RECORDS	in bytes
-	if (SWIZZLE_EN)
-		s_add_u32		s_save_buf_rsrc2, s_save_buf_rsrc2, 0x0						//FIXME need to use swizzle to enable bounds checking?
-	else
-		s_mov_b32		s_save_buf_rsrc2,  0x1000000								//NUM_RECORDS in bytes
-	end
-    s_and_b32       m0, s_wave_size, 1
-    s_cmp_eq_u32    m0, 1
-    s_cbranch_scc1  L_SAVE_DDID_WAVE64
-
-    write_sgpr_to_mem_wave32(s_save_m0, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF) 
-
-  L_SAVE_DDID_WAVE64:
-    write_sgpr_to_mem_wave64(s_save_m0, s_save_buf_rsrc0, s_save_mem_offset, SGPR_SAVE_USE_SQC, USE_MTBUF_INSTEAD_OF_MUBUF) 
-
-    end
-   
-  L_S_PGM_END_SAVED:
-	/*     S_PGM_END_SAVED  */    							//FIXME  graphics ONLY
-	if ((EMU_RUN_HACK) && (!EMU_RUN_HACK_SAVE_NORMAL_EXIT))	
-		s_and_b32 s_save_pc_hi, s_save_pc_hi, 0x0000ffff    //pc[47:32]
-		s_add_u32 s_save_pc_lo, s_save_pc_lo, 4             //pc[31:0]+4
-		s_addc_u32 s_save_pc_hi, s_save_pc_hi, 0x0			//carry bit over
-		s_rfe_b64 s_save_pc_lo                              //Return to the main shader program
-	else
-	end
-
-	
-    s_branch	L_END_PGM
-	
-
-				
-/**************************************************************************/
-/*                     	restore routine						              */
-/**************************************************************************/
+
+	/* inform SPI the readiness and wait for SPI's go signal */
+	s_mov_b32	s_save_exec_lo, exec_lo					//save EXEC and use EXEC for the go signal from SPI
+	s_mov_b32	s_save_exec_hi, exec_hi
+	s_mov_b64	exec, 0x0						//clear EXEC to get ready to receive
+
+	s_sendmsg	sendmsg(MSG_SAVEWAVE)					//send SPI a message and wait for SPI's write to EXEC
+
+L_SLEEP:
+	// sleep 1 (64clk) is not enough for 8 waves per SIMD, which will cause
+	// SQ hang, since the 7,8th wave could not get arbit to exec inst, while
+	// other waves are stuck into the sleep-loop and waiting for wrexec!=0
+	s_sleep		0x2
+	s_cbranch_execz	L_SLEEP
+
+	/* setup Resource Contants */
+	s_mov_b32	s_save_buf_rsrc0, s_save_spi_init_lo			//base_addr_lo
+	s_and_b32	s_save_buf_rsrc1, s_save_spi_init_hi, 0x0000FFFF	//base_addr_hi
+	s_or_b32	s_save_buf_rsrc1, s_save_buf_rsrc1, S_SAVE_BUF_RSRC_WORD1_STRIDE
+	s_mov_b32	s_save_buf_rsrc2, 0					//NUM_RECORDS initial value = 0 (in bytes) although not neccessarily inited
+	s_mov_b32	s_save_buf_rsrc3, S_SAVE_BUF_RSRC_WORD3_MISC
+	s_and_b32	s_save_tmp, s_save_spi_init_hi, S_SAVE_SPI_INIT_ATC_MASK
+	s_lshr_b32	s_save_tmp, s_save_tmp, (S_SAVE_SPI_INIT_ATC_SHIFT-SQ_BUF_RSRC_WORD1_ATC_SHIFT)
+	s_or_b32	s_save_buf_rsrc3, s_save_buf_rsrc3, s_save_tmp		//or ATC
+	s_and_b32	s_save_tmp, s_save_spi_init_hi, S_SAVE_SPI_INIT_MTYPE_MASK
+	s_lshr_b32	s_save_tmp, s_save_tmp, (S_SAVE_SPI_INIT_MTYPE_SHIFT-SQ_BUF_RSRC_WORD3_MTYPE_SHIFT)
+	s_or_b32	s_save_buf_rsrc3, s_save_buf_rsrc3, s_save_tmp		//or MTYPE
+
+	s_mov_b32	s_save_m0, m0
+
+	/* global mem offset */
+	s_mov_b32	s_save_mem_offset, 0x0
+	s_getreg_b32	s_wave_size, hwreg(HW_REG_IB_STS2,SQ_WAVE_IB_STS2_WAVE64_SHIFT,SQ_WAVE_IB_STS2_WAVE64_SIZE)
+	s_lshl_b32	s_wave_size, s_wave_size, S_WAVE_SIZE
+	s_or_b32	s_wave_size, s_save_spi_init_hi, s_wave_size		//share s_wave_size with exec_hi, it's at bit25
+
+	/* save HW registers */
+
+L_SAVE_HWREG:
+	// HWREG SR memory offset : size(VGPR)+size(SVGPR)+size(SGPR)
+	get_vgpr_size_bytes(s_save_mem_offset, s_wave_size)
+	get_svgpr_size_bytes(s_save_tmp)
+	s_add_u32	s_save_mem_offset, s_save_mem_offset, s_save_tmp
+	s_add_u32	s_save_mem_offset, s_save_mem_offset, get_sgpr_size_bytes()
+
+	s_mov_b32	s_save_buf_rsrc2, 0x1000000				//NUM_RECORDS in bytes
+
+	write_hwreg_to_mem(s_save_m0, s_save_buf_rsrc0, s_save_mem_offset)
+	write_hwreg_to_mem(s_save_pc_lo, s_save_buf_rsrc0, s_save_mem_offset)
+	write_hwreg_to_mem(s_save_pc_hi, s_save_buf_rsrc0, s_save_mem_offset)
+	write_hwreg_to_mem(s_save_exec_lo, s_save_buf_rsrc0, s_save_mem_offset)
+	write_hwreg_to_mem(s_save_exec_hi, s_save_buf_rsrc0, s_save_mem_offset)
+	write_hwreg_to_mem(s_save_status, s_save_buf_rsrc0, s_save_mem_offset)
+
+	s_getreg_b32	s_save_trapsts, hwreg(HW_REG_TRAPSTS)
+	write_hwreg_to_mem(s_save_trapsts, s_save_buf_rsrc0, s_save_mem_offset)
+	write_hwreg_to_mem(s_save_xnack_mask, s_save_buf_rsrc0, s_save_mem_offset)
+
+	s_getreg_b32	s_save_m0, hwreg(HW_REG_MODE)
+	write_hwreg_to_mem(s_save_m0, s_save_buf_rsrc0, s_save_mem_offset)
+
+	s_getreg_b32	s_save_m0, hwreg(HW_REG_SHADER_FLAT_SCRATCH_LO)
+	write_hwreg_to_mem(s_save_m0, s_save_buf_rsrc0, s_save_mem_offset)
+
+	s_getreg_b32	s_save_m0, hwreg(HW_REG_SHADER_FLAT_SCRATCH_HI)
+	write_hwreg_to_mem(s_save_m0, s_save_buf_rsrc0, s_save_mem_offset)
+
+	/* the first wave in the threadgroup */
+	s_and_b32	s_save_tmp, s_save_spi_init_hi, S_SAVE_SPI_INIT_FIRST_WAVE_MASK
+	s_mov_b32	s_save_exec_hi, 0x0
+	s_or_b32	s_save_exec_hi, s_save_tmp, s_save_exec_hi		// save first wave bit to s_save_exec_hi.bits[26]
+
+	/* save SGPRs */
+	// Save SGPR before LDS save, then the s0 to s4 can be used during LDS save...
+
+	// SGPR SR memory offset : size(VGPR)+size(SVGPR)
+	get_vgpr_size_bytes(s_save_mem_offset, s_wave_size)
+	get_svgpr_size_bytes(s_save_tmp)
+	s_add_u32	s_save_mem_offset, s_save_mem_offset, s_save_tmp
+	s_mov_b32	s_save_buf_rsrc2, 0x1000000				//NUM_RECORDS in bytes
+
+	// backup s_save_buf_rsrc0,1 to s_save_pc_lo/hi, since write_16sgpr_to_mem function will change the rsrc0
+	s_mov_b32	s_save_xnack_mask, s_save_buf_rsrc0
+	s_add_u32	s_save_buf_rsrc0, s_save_buf_rsrc0, s_save_mem_offset
+	s_addc_u32	s_save_buf_rsrc1, s_save_buf_rsrc1, 0
+
+	s_mov_b32	m0, 0x0							//SGPR initial index value =0
+	s_nop		0x0							//Manually inserted wait states
+L_SAVE_SGPR_LOOP:
+	// SGPR is allocated in 16 SGPR granularity
+	s_movrels_b64	s0, s0							//s0 = s[0+m0], s1 = s[1+m0]
+	s_movrels_b64	s2, s2							//s2 = s[2+m0], s3 = s[3+m0]
+	s_movrels_b64	s4, s4							//s4 = s[4+m0], s5 = s[5+m0]
+	s_movrels_b64	s6, s6							//s6 = s[6+m0], s7 = s[7+m0]
+	s_movrels_b64	s8, s8							//s8 = s[8+m0], s9 = s[9+m0]
+	s_movrels_b64	s10, s10						//s10 = s[10+m0], s11 = s[11+m0]
+	s_movrels_b64	s12, s12						//s12 = s[12+m0], s13 = s[13+m0]
+	s_movrels_b64	s14, s14						//s14 = s[14+m0], s15 = s[15+m0]
+
+	write_16sgpr_to_mem(s0, s_save_buf_rsrc0, s_save_mem_offset)
+	s_add_u32	m0, m0, 16						//next sgpr index
+	s_cmp_lt_u32	m0, 96							//scc = (m0 < first 96 SGPR) ? 1 : 0
+	s_cbranch_scc1	L_SAVE_SGPR_LOOP					//first 96 SGPR save is complete?
+
+	//save the rest 12 SGPR
+	s_movrels_b64	s0, s0							//s0 = s[0+m0], s1 = s[1+m0]
+	s_movrels_b64	s2, s2							//s2 = s[2+m0], s3 = s[3+m0]
+	s_movrels_b64	s4, s4							//s4 = s[4+m0], s5 = s[5+m0]
+	s_movrels_b64	s6, s6							//s6 = s[6+m0], s7 = s[7+m0]
+	s_movrels_b64	s8, s8							//s8 = s[8+m0], s9 = s[9+m0]
+	s_movrels_b64	s10, s10						//s10 = s[10+m0], s11 = s[11+m0]
+	write_12sgpr_to_mem(s0, s_save_buf_rsrc0, s_save_mem_offset)
+
+	// restore s_save_buf_rsrc0,1
+	s_mov_b32	s_save_buf_rsrc0, s_save_xnack_mask
+
+	/* save first 4 VGPR, then LDS save could use   */
+	// each wave will alloc 4 vgprs at least...
+
+	s_mov_b32	s_save_mem_offset, 0
+ 	s_mov_b32	exec_lo, 0xFFFFFFFF					//need every thread from now on
+	s_lshr_b32	m0, s_wave_size, S_WAVE_SIZE
+	s_and_b32	m0, m0, 1
+	s_cmp_eq_u32	m0, 1
+	s_cbranch_scc1	L_ENABLE_SAVE_4VGPR_EXEC_HI
+	s_mov_b32	exec_hi, 0x00000000
+	s_branch	L_SAVE_4VGPR_WAVE32
+L_ENABLE_SAVE_4VGPR_EXEC_HI:
+	s_mov_b32	exec_hi, 0xFFFFFFFF
+	s_branch	L_SAVE_4VGPR_WAVE64
+L_SAVE_4VGPR_WAVE32:
+	s_mov_b32	s_save_buf_rsrc2, 0x1000000				//NUM_RECORDS in bytes
+
+	// VGPR Allocated in 4-GPR granularity
+
+	buffer_store_dword	v0, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1
+	buffer_store_dword	v1, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1 offset:128
+	buffer_store_dword	v2, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1 offset:128*2
+	buffer_store_dword	v3, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1 offset:128*3
+	s_branch	L_SAVE_LDS
+
+L_SAVE_4VGPR_WAVE64:
+	s_mov_b32	s_save_buf_rsrc2, 0x1000000				//NUM_RECORDS in bytes
+
+	// VGPR Allocated in 4-GPR granularity
+
+	buffer_store_dword	v0, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1
+	buffer_store_dword	v1, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1 offset:256
+	buffer_store_dword	v2, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1 offset:256*2
+	buffer_store_dword	v3, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1 offset:256*3
+
+	/* save LDS */
+
+L_SAVE_LDS:
+	// Change EXEC to all threads...
+	s_mov_b32	exec_lo, 0xFFFFFFFF					//need every thread from now on
+	s_lshr_b32	m0, s_wave_size, S_WAVE_SIZE
+	s_and_b32	m0, m0, 1
+	s_cmp_eq_u32	m0, 1
+	s_cbranch_scc1	L_ENABLE_SAVE_LDS_EXEC_HI
+	s_mov_b32	exec_hi, 0x00000000
+	s_branch	L_SAVE_LDS_NORMAL
+L_ENABLE_SAVE_LDS_EXEC_HI:
+	s_mov_b32	exec_hi, 0xFFFFFFFF
+L_SAVE_LDS_NORMAL:
+	s_getreg_b32	s_save_alloc_size, hwreg(HW_REG_LDS_ALLOC,SQ_WAVE_LDS_ALLOC_LDS_SIZE_SHIFT,SQ_WAVE_LDS_ALLOC_LDS_SIZE_SIZE)
+	s_and_b32	s_save_alloc_size, s_save_alloc_size, 0xFFFFFFFF	//lds_size is zero?
+	s_cbranch_scc0	L_SAVE_LDS_DONE						//no lds used? jump to L_SAVE_DONE
+
+	s_barrier								//LDS is used? wait for other waves in the same TG
+	s_and_b32	s_save_tmp, s_save_exec_hi, S_SAVE_SPI_INIT_FIRST_WAVE_MASK
+	s_cbranch_scc0	L_SAVE_LDS_DONE
+
+	// first wave do LDS save;
+
+	s_lshl_b32	s_save_alloc_size, s_save_alloc_size, 6			//LDS size in dwords = lds_size * 64dw
+	s_lshl_b32	s_save_alloc_size, s_save_alloc_size, 2			//LDS size in bytes
+	s_mov_b32	s_save_buf_rsrc2, s_save_alloc_size			//NUM_RECORDS in bytes
+
+	// LDS at offset: size(VGPR)+size(SVGPR)+SIZE(SGPR)+SIZE(HWREG)
+	//
+	get_vgpr_size_bytes(s_save_mem_offset, s_wave_size)
+	get_svgpr_size_bytes(s_save_tmp)
+	s_add_u32	s_save_mem_offset, s_save_mem_offset, s_save_tmp
+	s_add_u32	s_save_mem_offset, s_save_mem_offset, get_sgpr_size_bytes()
+	s_add_u32	s_save_mem_offset, s_save_mem_offset, get_hwreg_size_bytes()
+
+	s_mov_b32	s_save_buf_rsrc2, 0x1000000				//NUM_RECORDS in bytes
+
+	//load 0~63*4(byte address) to vgpr v0
+	v_mbcnt_lo_u32_b32	v0, -1, 0
+	v_mbcnt_hi_u32_b32	v0, -1, v0
+	v_mul_u32_u24	v0, 4, v0
+
+	s_lshr_b32	m0, s_wave_size, S_WAVE_SIZE
+	s_and_b32	m0, m0, 1
+	s_cmp_eq_u32	m0, 1
+	s_mov_b32	m0, 0x0
+	s_cbranch_scc1	L_SAVE_LDS_W64
+
+L_SAVE_LDS_W32:
+	s_mov_b32	s3, 128
+	s_nop		0
+	s_nop		0
+	s_nop		0
+L_SAVE_LDS_LOOP_W32:
+	ds_read_b32	v1, v0
+	s_waitcnt	0
+	buffer_store_dword	v1, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1
+
+	s_add_u32	m0, m0, s3						//every buffer_store_lds does 256 bytes
+	s_add_u32	s_save_mem_offset, s_save_mem_offset, s3
+	v_add_nc_u32	v0, v0, 128						//mem offset increased by 128 bytes
+	s_cmp_lt_u32	m0, s_save_alloc_size					//scc=(m0 < s_save_alloc_size) ? 1 : 0
+	s_cbranch_scc1	L_SAVE_LDS_LOOP_W32					//LDS save is complete?
+
+	s_branch	L_SAVE_LDS_DONE
+
+L_SAVE_LDS_W64:
+	s_mov_b32	s3, 256
+	s_nop		0
+	s_nop		0
+	s_nop		0
+L_SAVE_LDS_LOOP_W64:
+	ds_read_b32	v1, v0
+	s_waitcnt	0
+	buffer_store_dword	v1, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1
+
+	s_add_u32	m0, m0, s3						//every buffer_store_lds does 256 bytes
+	s_add_u32	s_save_mem_offset, s_save_mem_offset, s3
+	v_add_nc_u32	v0, v0, 256						//mem offset increased by 256 bytes
+	s_cmp_lt_u32	m0, s_save_alloc_size					//scc=(m0 < s_save_alloc_size) ? 1 : 0
+	s_cbranch_scc1	L_SAVE_LDS_LOOP_W64					//LDS save is complete?
+
+L_SAVE_LDS_DONE:
+	/* save VGPRs  - set the Rest VGPRs */
+L_SAVE_VGPR:
+	// VGPR SR memory offset: 0
+	s_mov_b32	exec_lo, 0xFFFFFFFF					//need every thread from now on
+	s_lshr_b32	m0, s_wave_size, S_WAVE_SIZE
+	s_and_b32	m0, m0, 1
+	s_cmp_eq_u32	m0, 1
+	s_cbranch_scc1	L_ENABLE_SAVE_VGPR_EXEC_HI
+	s_mov_b32	s_save_mem_offset, (0+128*4)				// for the rest VGPRs
+	s_mov_b32	exec_hi, 0x00000000
+	s_branch	L_SAVE_VGPR_NORMAL
+L_ENABLE_SAVE_VGPR_EXEC_HI:
+	s_mov_b32	s_save_mem_offset, (0+256*4)				// for the rest VGPRs
+	s_mov_b32	exec_hi, 0xFFFFFFFF
+L_SAVE_VGPR_NORMAL:
+	s_getreg_b32	s_save_alloc_size, hwreg(HW_REG_GPR_ALLOC,SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SHIFT,SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SIZE)
+	s_add_u32	s_save_alloc_size, s_save_alloc_size, 1
+	s_lshl_b32	s_save_alloc_size, s_save_alloc_size, 2			//Number of VGPRs = (vgpr_size + 1) * 4    (non-zero value)
+	//determine it is wave32 or wave64
+	s_lshr_b32	m0, s_wave_size, S_WAVE_SIZE
+	s_and_b32	m0, m0, 1
+	s_cmp_eq_u32	m0, 1
+	s_cbranch_scc1	L_SAVE_VGPR_WAVE64
+
+	s_mov_b32	s_save_buf_rsrc2, 0x1000000				//NUM_RECORDS in bytes
+
+	// VGPR Allocated in 4-GPR granularity
+
+	// VGPR store using dw burst
+	s_mov_b32	m0, 0x4							//VGPR initial index value =4
+	s_cmp_lt_u32	m0, s_save_alloc_size
+	s_cbranch_scc0	L_SAVE_VGPR_END
+
+L_SAVE_VGPR_W32_LOOP:
+	v_movrels_b32	v0, v0							//v0 = v[0+m0]
+	v_movrels_b32	v1, v1							//v1 = v[1+m0]
+	v_movrels_b32	v2, v2							//v2 = v[2+m0]
+	v_movrels_b32	v3, v3							//v3 = v[3+m0]
+
+	buffer_store_dword	v0, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1
+	buffer_store_dword	v1, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1 offset:128
+	buffer_store_dword	v2, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1 offset:128*2
+	buffer_store_dword	v3, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1 offset:128*3
+
+	s_add_u32	m0, m0, 4						//next vgpr index
+	s_add_u32	s_save_mem_offset, s_save_mem_offset, 128*4		//every buffer_store_dword does 128 bytes
+	s_cmp_lt_u32	m0, s_save_alloc_size					//scc = (m0 < s_save_alloc_size) ? 1 : 0
+	s_cbranch_scc1	L_SAVE_VGPR_W32_LOOP					//VGPR save is complete?
+
+	s_branch	L_SAVE_VGPR_END
+
+L_SAVE_VGPR_WAVE64:
+	s_mov_b32	s_save_buf_rsrc2, 0x1000000				//NUM_RECORDS in bytes
+
+	// VGPR store using dw burst
+	s_mov_b32	m0, 0x4							//VGPR initial index value =4
+	s_cmp_lt_u32	m0, s_save_alloc_size
+	s_cbranch_scc0	L_SAVE_VGPR_END
+
+L_SAVE_VGPR_W64_LOOP:
+	v_movrels_b32	v0, v0							//v0 = v[0+m0]
+	v_movrels_b32	v1, v1							//v1 = v[1+m0]
+	v_movrels_b32	v2, v2							//v2 = v[2+m0]
+	v_movrels_b32	v3, v3							//v3 = v[3+m0]
+
+	buffer_store_dword	v0, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1
+	buffer_store_dword	v1, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1 offset:256
+	buffer_store_dword	v2, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1 offset:256*2
+	buffer_store_dword	v3, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1 offset:256*3
+
+	s_add_u32	m0, m0, 4						//next vgpr index
+	s_add_u32	s_save_mem_offset, s_save_mem_offset, 256*4		//every buffer_store_dword does 256 bytes
+	s_cmp_lt_u32	m0, s_save_alloc_size					//scc = (m0 < s_save_alloc_size) ? 1 : 0
+	s_cbranch_scc1	L_SAVE_VGPR_W64_LOOP					//VGPR save is complete?
+
+	//Below part will be the save shared vgpr part (new for gfx10)
+	s_getreg_b32	s_save_alloc_size, hwreg(HW_REG_LDS_ALLOC,SQ_WAVE_LDS_ALLOC_VGPR_SHARED_SIZE_SHIFT,SQ_WAVE_LDS_ALLOC_VGPR_SHARED_SIZE_SIZE)
+	s_and_b32	s_save_alloc_size, s_save_alloc_size, 0xFFFFFFFF	//shared_vgpr_size is zero?
+	s_cbranch_scc0	L_SAVE_VGPR_END						//no shared_vgpr used? jump to L_SAVE_LDS
+	s_lshl_b32	s_save_alloc_size, s_save_alloc_size, 3			//Number of SHARED_VGPRs = shared_vgpr_size * 8    (non-zero value)
+	//m0 now has the value of normal vgpr count, just add the m0 with shared_vgpr count to get the total count.
+	//save shared_vgpr will start from the index of m0
+	s_add_u32	s_save_alloc_size, s_save_alloc_size, m0
+	s_mov_b32	exec_lo, 0xFFFFFFFF
+	s_mov_b32	exec_hi, 0x00000000
+L_SAVE_SHARED_VGPR_WAVE64_LOOP:
+	v_movrels_b32	v0, v0							//v0 = v[0+m0]
+	buffer_store_dword	v0, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1
+	s_add_u32	m0, m0, 1						//next vgpr index
+	s_add_u32	s_save_mem_offset, s_save_mem_offset, 128
+	s_cmp_lt_u32	m0, s_save_alloc_size					//scc = (m0 < s_save_alloc_size) ? 1 : 0
+	s_cbranch_scc1	L_SAVE_SHARED_VGPR_WAVE64_LOOP				//SHARED_VGPR save is complete?
+
+L_SAVE_VGPR_END:
+	s_branch	L_END_PGM
 
 L_RESTORE:
-    /*      Setup Resource Contants    */
-    if ((EMU_RUN_HACK) && (!EMU_RUN_HACK_RESTORE_NORMAL))
-		//calculate wd_addr using absolute thread id
-		v_readlane_b32 s_restore_tmp, v9, 0
-        //determine it is wave32 or wave64
-        s_getreg_b32 	s_restore_size, hwreg(HW_REG_IB_STS2,SQ_WAVE_IB_STS2_WAVE64_SHIFT,SQ_WAVE_IB_STS2_WAVE64_SIZE) //change to ttmp13
-        s_cmp_eq_u32    s_restore_size, 0
-        s_cbranch_scc1  L_RESTORE_WAVE32
-        s_lshr_b32 s_restore_tmp, s_restore_tmp, 6 //SAVE WAVE64
-        s_branch    L_RESTORE_CON
-    L_RESTORE_WAVE32:
-        s_lshr_b32 s_restore_tmp, s_restore_tmp, 5 //SAVE WAVE32
-    L_RESTORE_CON:
-		s_mul_i32 s_restore_tmp, s_restore_tmp, WAVE_SPACE
-		s_add_i32 s_restore_spi_init_lo, s_restore_tmp, WG_BASE_ADDR_LO
-		s_mov_b32 s_restore_spi_init_hi, WG_BASE_ADDR_HI
-		s_and_b32 s_restore_spi_init_hi, s_restore_spi_init_hi, CTX_RESTORE_CONTROL	
-	else
-	end
-	
-    s_mov_b32		s_restore_buf_rsrc0, 	s_restore_spi_init_lo															//base_addr_lo
-	s_and_b32		s_restore_buf_rsrc1, 	s_restore_spi_init_hi, 0x0000FFFF												//base_addr_hi
-	s_or_b32		s_restore_buf_rsrc1, 	s_restore_buf_rsrc1,  S_RESTORE_BUF_RSRC_WORD1_STRIDE
-    s_mov_b32       s_restore_buf_rsrc2,   	0                                               								//NUM_RECORDS initial value = 0 (in bytes)
-	s_mov_b32		s_restore_buf_rsrc3, 	S_RESTORE_BUF_RSRC_WORD3_MISC
-	s_and_b32		s_restore_tmp,         	s_restore_spi_init_hi, S_RESTORE_SPI_INIT_ATC_MASK		
-	s_lshr_b32		s_restore_tmp,  		s_restore_tmp, (S_RESTORE_SPI_INIT_ATC_SHIFT-SQ_BUF_RSRC_WORD1_ATC_SHIFT)		//get ATC bit into position
-	s_or_b32		s_restore_buf_rsrc3, 	s_restore_buf_rsrc3,  s_restore_tmp												//or ATC
-	s_and_b32		s_restore_tmp,         	s_restore_spi_init_hi, S_RESTORE_SPI_INIT_MTYPE_MASK		
-	s_lshr_b32		s_restore_tmp,  		s_restore_tmp, (S_RESTORE_SPI_INIT_MTYPE_SHIFT-SQ_BUF_RSRC_WORD3_MTYPE_SHIFT)	//get MTYPE bits into position
-	s_or_b32		s_restore_buf_rsrc3, 	s_restore_buf_rsrc3,  s_restore_tmp												//or MTYPE
-    //determine it is wave32 or wave64
-    s_getreg_b32 	s_restore_size, hwreg(HW_REG_IB_STS2,SQ_WAVE_IB_STS2_WAVE64_SHIFT,SQ_WAVE_IB_STS2_WAVE64_SIZE)
-    s_or_b32        s_restore_size, s_restore_spi_init_hi,    s_restore_size                                             //share s_wave_size with exec_hi
-	
-	/* 		global mem offset			*/
-	s_mov_b32		s_restore_mem_offset, 0x0								//mem offset initial value = 0
-
-        /*      	restore VGPRs	    */
-	//////////////////////////////
-  L_RESTORE_VGPR:
-  
- 	s_mov_b32		exec_lo, 0xFFFFFFFF 													//need every thread from now on   //be consistent with SAVE although can be moved ahead
-    s_and_b32       m0, s_restore_size, 1
-    s_cmp_eq_u32    m0, 1
-    s_cbranch_scc1  L_ENABLE_RESTORE_VGPR_EXEC_HI   
-    s_mov_b32		exec_hi, 0x00000000
-    s_branch        L_RESTORE_VGPR_NORMAL
-  L_ENABLE_RESTORE_VGPR_EXEC_HI:
-	s_mov_b32		exec_hi, 0xFFFFFFFF
-  L_RESTORE_VGPR_NORMAL:	
-	s_getreg_b32 	s_restore_alloc_size, hwreg(HW_REG_GPR_ALLOC,SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SHIFT,SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SIZE) 	//vpgr_size
-	s_add_u32 		s_restore_alloc_size, s_restore_alloc_size, 1
-	s_lshl_b32 		s_restore_alloc_size, s_restore_alloc_size, 2 							//Number of VGPRs = (vgpr_size + 1) * 4    (non-zero value)
-    //determine it is wave32 or wave64
-    s_and_b32       m0, s_restore_size, 1
-    s_cmp_eq_u32    m0, 1
-    s_cbranch_scc1  L_RESTORE_VGPR_WAVE64
-
-    s_lshl_b32		s_restore_buf_rsrc2,  s_restore_alloc_size, 7						    //NUM_RECORDS in bytes (32 threads*4)
-	if (SWIZZLE_EN)
-		s_add_u32		s_restore_buf_rsrc2, s_restore_buf_rsrc2, 0x0						//FIXME need to use swizzle to enable bounds checking?
-	else
-		s_mov_b32		s_restore_buf_rsrc2,  0x1000000										//NUM_RECORDS in bytes
-	end	
-
-	s_mov_b32		s_restore_mem_offset_save, s_restore_mem_offset							// restore start with v1, v0 will be the last
-	s_add_u32		s_restore_mem_offset, s_restore_mem_offset, 128
-    s_mov_b32 		m0, 1 																	//VGPR initial index value = 1
-	//s_set_gpr_idx_on  m0, 0x8																//M0[7:0] = M0[7:0] and M0[15:12] = 0x8
-    //s_add_u32		s_restore_alloc_size, s_restore_alloc_size, 0x8000						//add 0x8000 since we compare m0 against it later, might not need this in gfx10	
-
-  L_RESTORE_VGPR_WAVE32_LOOP: 										
-    if(USE_MTBUF_INSTEAD_OF_MUBUF)       
-		tbuffer_load_format_x v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset format:BUF_NUM_FORMAT_FLOAT format: BUF_DATA_FORMAT_32 slc:1 glc:1
-    else
-		buffer_load_dword v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset	slc:1 glc:1	
-	end
-	s_waitcnt		vmcnt(0)																//ensure data ready
-	v_movreld_b32		v0, v0																	//v[0+m0] = v0
-    s_add_u32		m0, m0, 1																//next vgpr index
-	s_add_u32		s_restore_mem_offset, s_restore_mem_offset, 128							//every buffer_load_dword does 128 bytes
-	s_cmp_lt_u32 	m0,	s_restore_alloc_size 												//scc = (m0 < s_restore_alloc_size) ? 1 : 0
-	s_cbranch_scc1 	L_RESTORE_VGPR_WAVE32_LOOP														//VGPR restore (except v0) is complete?
-	//s_set_gpr_idx_off
-																							/* VGPR restore on v0 */
-    if(USE_MTBUF_INSTEAD_OF_MUBUF)       
-		tbuffer_load_format_x v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save format:BUF_NUM_FORMAT_FLOAT format: BUF_DATA_FORMAT_32 slc:1 glc:1
-    else
-		buffer_load_dword v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save	slc:1 glc:1	
-	end
-
-    s_branch    L_RESTORE_LDS
-
-  L_RESTORE_VGPR_WAVE64:
-    s_lshl_b32		s_restore_buf_rsrc2,  s_restore_alloc_size, 8						    //NUM_RECORDS in bytes (64 threads*4)
-	if (SWIZZLE_EN)
-		s_add_u32		s_restore_buf_rsrc2, s_restore_buf_rsrc2, 0x0						//FIXME need to use swizzle to enable bounds checking?
-	else
-		s_mov_b32		s_restore_buf_rsrc2,  0x1000000										//NUM_RECORDS in bytes
-	end	
-
-	s_mov_b32		s_restore_mem_offset_save, s_restore_mem_offset							// restore start with v1, v0 will be the last
-	s_add_u32		s_restore_mem_offset, s_restore_mem_offset, 256
-    s_mov_b32 		m0, 1 																	//VGPR initial index value = 1
-  L_RESTORE_VGPR_WAVE64_LOOP: 										
-    if(USE_MTBUF_INSTEAD_OF_MUBUF)       
-		tbuffer_load_format_x v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset format:BUF_NUM_FORMAT_FLOAT format: BUF_DATA_FORMAT_32 slc:1 glc:1
-    else
-		buffer_load_dword v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset	slc:1 glc:1	
-	end
-	s_waitcnt		vmcnt(0)																//ensure data ready
-	v_movreld_b32		v0, v0																	//v[0+m0] = v0
-    s_add_u32		m0, m0, 1																//next vgpr index
-	s_add_u32		s_restore_mem_offset, s_restore_mem_offset, 256							//every buffer_load_dword does 256 bytes
-	s_cmp_lt_u32 	m0,	s_restore_alloc_size 												//scc = (m0 < s_restore_alloc_size) ? 1 : 0
-	s_cbranch_scc1 	L_RESTORE_VGPR_WAVE64_LOOP														//VGPR restore (except v0) is complete?
-	//s_set_gpr_idx_off
-    //
-    //Below part will be the restore shared vgpr part (new for gfx10)
-    s_getreg_b32 	s_restore_alloc_size, hwreg(HW_REG_LDS_ALLOC,SQ_WAVE_LDS_ALLOC_VGPR_SHARED_SIZE_SHIFT,SQ_WAVE_LDS_ALLOC_VGPR_SHARED_SIZE_SIZE) 			//shared_vgpr_size
-    s_and_b32		s_restore_alloc_size, s_restore_alloc_size, 0xFFFFFFFF				//shared_vgpr_size is zero?
-    s_cbranch_scc0	L_RESTORE_V0													    //no shared_vgpr used? jump to L_SAVE_LDS
-    s_lshl_b32 		s_restore_alloc_size, s_restore_alloc_size, 3 						//Number of SHARED_VGPRs = shared_vgpr_size * 8    (non-zero value)
-    //m0 now has the value of normal vgpr count, just add the m0 with shared_vgpr count to get the total count.
-    //restore shared_vgpr will start from the index of m0
-    s_add_u32       s_restore_alloc_size, s_restore_alloc_size, m0
-    s_mov_b32		exec_lo, 0xFFFFFFFF
-    s_mov_b32		exec_hi, 0x00000000
-    L_RESTORE_SHARED_VGPR_WAVE64_LOOP: 
-    buffer_load_dword v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset	slc:1 glc:1
-    s_waitcnt		vmcnt(0)																//ensure data ready
-	v_movreld_b32		v0, v0																	//v[0+m0] = v0
-    s_add_u32		m0, m0, 1																//next vgpr index
-	s_add_u32		s_restore_mem_offset, s_restore_mem_offset, 128							//every buffer_load_dword does 256 bytes
-	s_cmp_lt_u32 	m0,	s_restore_alloc_size 												//scc = (m0 < s_restore_alloc_size) ? 1 : 0
-	s_cbranch_scc1 	L_RESTORE_SHARED_VGPR_WAVE64_LOOP														//VGPR restore (except v0) is complete?
-
-    s_mov_b32 exec_hi, 0xFFFFFFFF                                                           //restore back exec_hi before restoring V0!!
-	
-    /* VGPR restore on v0 */
-  L_RESTORE_V0:
-    if(USE_MTBUF_INSTEAD_OF_MUBUF)       
-		tbuffer_load_format_x v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save format:BUF_NUM_FORMAT_FLOAT format: BUF_DATA_FORMAT_32 slc:1 glc:1
-    else
-		buffer_load_dword v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save	slc:1 glc:1	
-	end
-
-
-    /*      	restore LDS	    */
-	//////////////////////////////
-  L_RESTORE_LDS:
-
-    //Only need to check the first wave    
-	/*      the first wave in the threadgroup    */
-	s_and_b32		s_restore_tmp, s_restore_size, S_RESTORE_SPI_INIT_FIRST_WAVE_MASK			
-	s_cbranch_scc0	L_RESTORE_SGPR
-	
-    s_mov_b32		exec_lo, 0xFFFFFFFF 													//need every thread from now on   //be consistent with SAVE although can be moved ahead
-    s_and_b32       m0, s_restore_size, 1
-    s_cmp_eq_u32    m0, 1
-    s_cbranch_scc1  L_ENABLE_RESTORE_LDS_EXEC_HI   
-    s_mov_b32		exec_hi, 0x00000000
-    s_branch        L_RESTORE_LDS_NORMAL
-  L_ENABLE_RESTORE_LDS_EXEC_HI:
-	s_mov_b32		exec_hi, 0xFFFFFFFF
-  L_RESTORE_LDS_NORMAL:	
-	s_getreg_b32 	s_restore_alloc_size, hwreg(HW_REG_LDS_ALLOC,SQ_WAVE_LDS_ALLOC_LDS_SIZE_SHIFT,SQ_WAVE_LDS_ALLOC_LDS_SIZE_SIZE) 				//lds_size
-	s_and_b32		s_restore_alloc_size, s_restore_alloc_size, 0xFFFFFFFF					//lds_size is zero?
-	s_cbranch_scc0	L_RESTORE_SGPR															//no lds used? jump to L_RESTORE_VGPR
-	s_lshl_b32 		s_restore_alloc_size, s_restore_alloc_size, 6 							//LDS size in dwords = lds_size * 64dw
-	s_lshl_b32 		s_restore_alloc_size, s_restore_alloc_size, 2 							//LDS size in bytes
-	s_mov_b32		s_restore_buf_rsrc2,	s_restore_alloc_size							//NUM_RECORDS in bytes
-	if (SWIZZLE_EN)
-		s_add_u32		s_restore_buf_rsrc2, s_restore_buf_rsrc2, 0x0						//FIXME need to use swizzle to enable bounds checking?
-	else
-		s_mov_b32		s_restore_buf_rsrc2,  0x1000000										//NUM_RECORDS in bytes
-	end
-
-    s_and_b32       m0, s_wave_size, 1
-    s_cmp_eq_u32    m0, 1
-    s_mov_b32 		m0, 0x0
-    s_cbranch_scc1  L_RESTORE_LDS_LOOP_W64
-
-  L_RESTORE_LDS_LOOP_W32:									
-	if (SAVE_LDS)
-	buffer_load_dword	v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset lds:1
-    s_waitcnt 0
-	end
-    s_add_u32		m0, m0, 128																//every buffer_load_dword does 256 bytes
-	s_add_u32		s_restore_mem_offset, s_restore_mem_offset, 128						//mem offset increased by 256 bytes
-	s_cmp_lt_u32	m0, s_restore_alloc_size												//scc=(m0 < s_restore_alloc_size) ? 1 : 0
-	s_cbranch_scc1  L_RESTORE_LDS_LOOP_W32														//LDS restore is complete?
-    s_branch        L_RESTORE_SGPR
-
-  L_RESTORE_LDS_LOOP_W64:									
-	if (SAVE_LDS)
-	buffer_load_dword	v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset lds:1
-    s_waitcnt 0
-	end
-    s_add_u32		m0, m0, 256																//every buffer_load_dword does 256 bytes
-	s_add_u32		s_restore_mem_offset, s_restore_mem_offset, 256							//mem offset increased by 256 bytes
-	s_cmp_lt_u32	m0, s_restore_alloc_size												//scc=(m0 < s_restore_alloc_size) ? 1 : 0
-	s_cbranch_scc1  L_RESTORE_LDS_LOOP_W64														//LDS restore is complete?
-
-	
-    /*      	restore SGPRs	    */
-	//////////////////////////////
-	//s_getreg_b32 	s_restore_alloc_size, hwreg(HW_REG_GPR_ALLOC,SQ_WAVE_GPR_ALLOC_SGPR_SIZE_SHIFT,SQ_WAVE_GPR_ALLOC_SGPR_SIZE_SIZE) 				//spgr_size
-	//s_add_u32 		s_restore_alloc_size, s_restore_alloc_size, 1
-	//s_lshl_b32 		s_restore_alloc_size, s_restore_alloc_size, 4 							//Number of SGPRs = (sgpr_size + 1) * 16   (non-zero value)
-	//s_lshl_b32 		s_restore_alloc_size, s_restore_alloc_size, 3 							//Number of SGPRs = (sgpr_size + 1) * 8   (non-zero value)
-  L_RESTORE_SGPR:
-    //need to look at it is wave32 or wave64
-    s_and_b32       m0, s_restore_size, 1
-    s_cmp_eq_u32    m0, 1
-    s_cbranch_scc1  L_RESTORE_SGPR_VMEM_WAVE64
-	if (SGPR_SAVE_USE_SQC)
-		s_lshl_b32		s_restore_buf_rsrc2,	s_sgpr_save_num, 2						//NUM_RECORDS in bytes 
-	else
-        s_lshl_b32		s_restore_buf_rsrc2,	s_sgpr_save_num, 7						//NUM_RECORDS in bytes (32 threads)
-    end
-    s_branch        L_RESTORE_SGPR_CONT
-  L_RESTORE_SGPR_VMEM_WAVE64:
-    if (SGPR_SAVE_USE_SQC)
-		s_lshl_b32		s_restore_buf_rsrc2,	s_sgpr_save_num, 2						//NUM_RECORDS in bytes 
-	else
-		s_lshl_b32		s_restore_buf_rsrc2,	s_sgpr_save_num, 8						//NUM_RECORDS in bytes (64 threads)
-	end
-
-  L_RESTORE_SGPR_CONT:
-	if (SWIZZLE_EN)
-		s_add_u32		s_restore_buf_rsrc2, s_restore_buf_rsrc2, 0x0						//FIXME need to use swizzle to enable bounds checking?
-	else
-		s_mov_b32		s_restore_buf_rsrc2,  0x1000000										//NUM_RECORDS in bytes
-	end
-
-    s_and_b32       m0, s_restore_size, 1
-    s_cmp_eq_u32    m0, 1
-    s_cbranch_scc1  L_RESTORE_SGPR_WAVE64
-
-    read_sgpr_from_mem_wave32(s_restore_tmp, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)		//save s0 to s_restore_tmp
-	s_mov_b32 		m0, 0x1
-
-  L_RESTORE_SGPR_LOOP_WAVE32:
-    read_sgpr_from_mem_wave32(s0, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)															//PV: further performance improvement can be made
-	s_waitcnt		lgkmcnt(0)																//ensure data ready
-	s_movreld_b32 	s0, s0                                                                  //s[0+m0] = s0
-    s_nop 0                                                                                 // hazard SALU M0=> S_MOVREL
-	s_add_u32		m0, m0, 1																//next sgpr index
-	s_cmp_lt_u32 	m0, s_sgpr_save_num												//scc = (m0 < s_restore_alloc_size) ? 1 : 0
-	s_cbranch_scc1 	L_RESTORE_SGPR_LOOP_WAVE32														//SGPR restore (except s0) is complete?
-	s_mov_b32		s0, s_restore_tmp															/* SGPR restore on s0 */
-    s_branch        L_RESTORE_HWREG
-  
-  L_RESTORE_SGPR_WAVE64:
-	read_sgpr_from_mem_wave64(s_restore_tmp, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)		//save s0 to s_restore_tmp
-	s_mov_b32 		m0, 0x1																				//SGPR initial index value =1	//go on with with s1
-	
-  L_RESTORE_SGPR_LOOP_WAVE64: 																					
-	read_sgpr_from_mem_wave64(s0, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)															//PV: further performance improvement can be made
-	s_waitcnt		lgkmcnt(0)																//ensure data ready
-	s_movreld_b32 	s0, s0                                                                  //s[0+m0] = s0
-    s_nop 0                                                                                 // hazard SALU M0=> S_MOVREL
-	s_add_u32		m0, m0, 1																//next sgpr index
-	s_cmp_lt_u32 	m0, s_sgpr_save_num												//scc = (m0 < s_restore_alloc_size) ? 1 : 0
-	s_cbranch_scc1 	L_RESTORE_SGPR_LOOP_WAVE64														//SGPR restore (except s0) is complete?
-	s_mov_b32		s0, s_restore_tmp															/* SGPR restore on s0 */
-
-	
-    /* 		restore HW registers	*/
-	//////////////////////////////
-  L_RESTORE_HWREG:
-    s_mov_b32		s_restore_buf_rsrc2, 0x4												//NUM_RECORDS	in bytes
-	if (SWIZZLE_EN)
-		s_add_u32		s_restore_buf_rsrc2, s_restore_buf_rsrc2, 0x0						//FIXME need to use swizzle to enable bounds checking?
-	else
-		s_mov_b32		s_restore_buf_rsrc2,  0x1000000										//NUM_RECORDS in bytes
-	end
-
-    s_and_b32       m0, s_restore_size, 1
-    s_cmp_eq_u32    m0, 1
-    s_cbranch_scc1  L_RESTORE_HWREG_WAVE64
-
-    read_sgpr_from_mem_wave32(s_restore_m0, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)					//M0
-	read_sgpr_from_mem_wave32(s_restore_pc_lo, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)				//PC
-	read_sgpr_from_mem_wave32(s_restore_pc_hi, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)
-	read_sgpr_from_mem_wave32(s_restore_exec_lo, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)				//EXEC
-	read_sgpr_from_mem_wave32(s_restore_exec_hi, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)
-	read_sgpr_from_mem_wave32(s_restore_status, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)				//STATUS
-	read_sgpr_from_mem_wave32(s_restore_trapsts, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)				//TRAPSTS
-    //read_sgpr_from_mem_wave32(xnack_mask_lo, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)					//XNACK_MASK_LO
-	//read_sgpr_from_mem_wave32(xnack_mask_hi, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)					//XNACK_MASK_HI
-    read_sgpr_from_mem_wave32(s_restore_xnack_mask, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)					//XNACK_MASK
-	read_sgpr_from_mem_wave32(s_restore_mode, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)				//MODE
-    if(SAVE_RESTORE_HWID_DDID)
-    read_sgpr_from_mem_wave32(s_restore_hwid1, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)				//HW_ID1
-    end
-    s_branch        L_RESTORE_HWREG_FINISH
-
-  L_RESTORE_HWREG_WAVE64:
-	read_sgpr_from_mem_wave64(s_restore_m0, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)					//M0
-	read_sgpr_from_mem_wave64(s_restore_pc_lo, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)				//PC
-	read_sgpr_from_mem_wave64(s_restore_pc_hi, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)
-	read_sgpr_from_mem_wave64(s_restore_exec_lo, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)				//EXEC
-	read_sgpr_from_mem_wave64(s_restore_exec_hi, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)
-	read_sgpr_from_mem_wave64(s_restore_status, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)				//STATUS
-	read_sgpr_from_mem_wave64(s_restore_trapsts, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)				//TRAPSTS
-    //read_sgpr_from_mem_wave64(xnack_mask_lo, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)					//XNACK_MASK_LO
-	//read_sgpr_from_mem_wave64(xnack_mask_hi, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)					//XNACK_MASK_HI
-    read_sgpr_from_mem_wave64(s_restore_xnack_mask, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)					//XNACK_MASK
-	read_sgpr_from_mem_wave64(s_restore_mode, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)				//MODE
-    if(SAVE_RESTORE_HWID_DDID)
-    read_sgpr_from_mem_wave64(s_restore_hwid1, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)				//HW_ID1
-    end
-  L_RESTORE_HWREG_FINISH:
-	s_waitcnt		lgkmcnt(0)																						//from now on, it is safe to restore STATUS and IB_STS
-  
-
-
-    if(SAVE_RESTORE_HWID_DDID)
-  L_RESTORE_DDID:
-    s_mov_b32      m0, s_restore_hwid1                                                      //virture ttrace support: The save-context handler records the SE/SA/WGP/SIMD/wave of the original wave
-    s_ttracedata                                                                            //and then can output it as SHADER_DATA to ttrace on restore to provide a correlation across the save-restore
-                                    
-    s_mov_b32		s_restore_buf_rsrc2, 0x4												//NUM_RECORDS	in bytes
-	if (SWIZZLE_EN)
-		s_add_u32		s_restore_buf_rsrc2, s_restore_buf_rsrc2, 0x0						//FIXME need to use swizzle to enable bounds checking?
-	else
-		s_mov_b32		s_restore_buf_rsrc2,  0x1000000										//NUM_RECORDS in bytes
-	end
-
-    s_and_b32       m0, s_restore_size, 1
-    s_cmp_eq_u32    m0, 1
-    s_cbranch_scc1  L_RESTORE_DDID_WAVE64
-
-    read_sgpr_from_mem_wave32(s_restore_ddid, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)	
-    s_branch        L_RESTORE_DDID_FINISH
-  L_RESTORE_DDID_WAVE64:
-    read_sgpr_from_mem_wave64(s_restore_ddid, s_restore_buf_rsrc0, s_restore_mem_offset, SGPR_SAVE_USE_SQC)	
-
-  L_RESTORE_DDID_FINISH:
-    s_waitcnt		lgkmcnt(0)
-    //s_mov_b32      m0, s_restore_ddid
-    //s_ttracedata   
-    if (RESTORE_DDID_IN_SGPR18)
-        s_mov_b32   s18, s_restore_ddid
-	end	
-    
-    end   
-
-	s_and_b32 s_restore_pc_hi, s_restore_pc_hi, 0x0000ffff    	//pc[47:32]        //Do it here in order not to affect STATUS
-
-	//for normal save & restore, the saved PC points to the next inst to execute, no adjustment needs to be made, otherwise:
-	if ((EMU_RUN_HACK) && (!EMU_RUN_HACK_RESTORE_NORMAL))
-		s_add_u32 s_restore_pc_lo, s_restore_pc_lo, 8            //pc[31:0]+8	  //two back-to-back s_trap are used (first for save and second for restore)
-		s_addc_u32	s_restore_pc_hi, s_restore_pc_hi, 0x0		 //carry bit over
-	end	
-	if ((EMU_RUN_HACK) && (EMU_RUN_HACK_RESTORE_NORMAL))	      
-		s_add_u32 s_restore_pc_lo, s_restore_pc_lo, 4            //pc[31:0]+4     // save is hack through s_trap but restore is normal
-		s_addc_u32	s_restore_pc_hi, s_restore_pc_hi, 0x0		 //carry bit over
-	end
-	
-	s_mov_b32 		m0, 		s_restore_m0
-	s_mov_b32 		exec_lo, 	s_restore_exec_lo
-	s_mov_b32 		exec_hi, 	s_restore_exec_hi
-	
-	s_and_b32		s_restore_m0, SQ_WAVE_TRAPSTS_PRE_SAVECTX_MASK, s_restore_trapsts
+	/* Setup Resource Contants */
+	s_mov_b32	s_restore_buf_rsrc0, s_restore_spi_init_lo		//base_addr_lo
+	s_and_b32	s_restore_buf_rsrc1, s_restore_spi_init_hi, 0x0000FFFF	//base_addr_hi
+	s_or_b32	s_restore_buf_rsrc1, s_restore_buf_rsrc1, S_RESTORE_BUF_RSRC_WORD1_STRIDE
+	s_mov_b32	s_restore_buf_rsrc2, 0					//NUM_RECORDS initial value = 0 (in bytes)
+	s_mov_b32	s_restore_buf_rsrc3, S_RESTORE_BUF_RSRC_WORD3_MISC
+	s_and_b32	s_restore_tmp, s_restore_spi_init_hi, S_RESTORE_SPI_INIT_ATC_MASK
+	s_lshr_b32	s_restore_tmp, s_restore_tmp, (S_RESTORE_SPI_INIT_ATC_SHIFT-SQ_BUF_RSRC_WORD1_ATC_SHIFT)
+	s_or_b32	s_restore_buf_rsrc3, s_restore_buf_rsrc3, s_restore_tmp	//or ATC
+	s_and_b32	s_restore_tmp, s_restore_spi_init_hi, S_RESTORE_SPI_INIT_MTYPE_MASK
+	s_lshr_b32	s_restore_tmp, s_restore_tmp, (S_RESTORE_SPI_INIT_MTYPE_SHIFT-SQ_BUF_RSRC_WORD3_MTYPE_SHIFT)
+	s_or_b32	s_restore_buf_rsrc3, s_restore_buf_rsrc3, s_restore_tmp	//or MTYPE
+	//determine it is wave32 or wave64
+	s_getreg_b32	s_restore_size, hwreg(HW_REG_IB_STS2,SQ_WAVE_IB_STS2_WAVE64_SHIFT,SQ_WAVE_IB_STS2_WAVE64_SIZE)
+	s_lshl_b32	s_restore_size, s_restore_size, S_WAVE_SIZE
+	s_or_b32	s_restore_size, s_restore_spi_init_hi, s_restore_size
+
+	s_and_b32	s_restore_tmp, s_restore_spi_init_hi, S_RESTORE_SPI_INIT_FIRST_WAVE_MASK
+	s_cbranch_scc0	L_RESTORE_VGPR
+
+	/* restore LDS */
+L_RESTORE_LDS:
+	s_mov_b32	exec_lo, 0xFFFFFFFF					//need every thread from now on
+	s_lshr_b32	m0, s_restore_size, S_WAVE_SIZE
+	s_and_b32	m0, m0, 1
+	s_cmp_eq_u32	m0, 1
+	s_cbranch_scc1	L_ENABLE_RESTORE_LDS_EXEC_HI
+	s_mov_b32	exec_hi, 0x00000000
+	s_branch	L_RESTORE_LDS_NORMAL
+L_ENABLE_RESTORE_LDS_EXEC_HI:
+	s_mov_b32	exec_hi, 0xFFFFFFFF
+L_RESTORE_LDS_NORMAL:
+	s_getreg_b32	s_restore_alloc_size, hwreg(HW_REG_LDS_ALLOC,SQ_WAVE_LDS_ALLOC_LDS_SIZE_SHIFT,SQ_WAVE_LDS_ALLOC_LDS_SIZE_SIZE)
+	s_and_b32	s_restore_alloc_size, s_restore_alloc_size, 0xFFFFFFFF	//lds_size is zero?
+	s_cbranch_scc0	L_RESTORE_VGPR						//no lds used? jump to L_RESTORE_VGPR
+	s_lshl_b32	s_restore_alloc_size, s_restore_alloc_size, 6		//LDS size in dwords = lds_size * 64dw
+	s_lshl_b32	s_restore_alloc_size, s_restore_alloc_size, 2		//LDS size in bytes
+	s_mov_b32	s_restore_buf_rsrc2, s_restore_alloc_size		//NUM_RECORDS in bytes
+
+	// LDS at offset: size(VGPR)+size(SVGPR)+SIZE(SGPR)+SIZE(HWREG)
+	//
+	get_vgpr_size_bytes(s_restore_mem_offset, s_restore_size)
+	get_svgpr_size_bytes(s_restore_tmp)
+	s_add_u32	s_restore_mem_offset, s_restore_mem_offset, s_restore_tmp
+	s_add_u32	s_restore_mem_offset, s_restore_mem_offset, get_sgpr_size_bytes()
+	s_add_u32	s_restore_mem_offset, s_restore_mem_offset, get_hwreg_size_bytes()
+
+	s_mov_b32	s_restore_buf_rsrc2, 0x1000000				//NUM_RECORDS in bytes
+
+	s_lshr_b32	m0, s_wave_size, S_WAVE_SIZE
+	s_and_b32	m0, m0, 1
+	s_cmp_eq_u32	m0, 1
+	s_mov_b32	m0, 0x0
+	s_cbranch_scc1	L_RESTORE_LDS_LOOP_W64
+
+L_RESTORE_LDS_LOOP_W32:
+	buffer_load_dword	v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset lds:1	// first 64DW
+	s_add_u32	m0, m0, 128						// 128 DW
+	s_add_u32	s_restore_mem_offset, s_restore_mem_offset, 128		//mem offset increased by 128DW
+	s_cmp_lt_u32	m0, s_restore_alloc_size				//scc=(m0 < s_restore_alloc_size) ? 1 : 0
+	s_cbranch_scc1	L_RESTORE_LDS_LOOP_W32					//LDS restore is complete?
+	s_branch	L_RESTORE_VGPR
+
+L_RESTORE_LDS_LOOP_W64:
+	buffer_load_dword	v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset lds:1	// first 64DW
+	s_add_u32	m0, m0, 256						// 256 DW
+	s_add_u32	s_restore_mem_offset, s_restore_mem_offset, 256		//mem offset increased by 256DW
+	s_cmp_lt_u32	m0, s_restore_alloc_size				//scc=(m0 < s_restore_alloc_size) ? 1 : 0
+	s_cbranch_scc1	L_RESTORE_LDS_LOOP_W64					//LDS restore is complete?
+
+	/* restore VGPRs */
+L_RESTORE_VGPR:
+	// VGPR SR memory offset : 0
+	s_mov_b32	s_restore_mem_offset, 0x0
+ 	s_mov_b32	exec_lo, 0xFFFFFFFF					//need every thread from now on
+	s_lshr_b32	m0, s_restore_size, S_WAVE_SIZE
+	s_and_b32	m0, m0, 1
+	s_cmp_eq_u32	m0, 1
+	s_cbranch_scc1	L_ENABLE_RESTORE_VGPR_EXEC_HI
+	s_mov_b32	exec_hi, 0x00000000
+	s_branch	L_RESTORE_VGPR_NORMAL
+L_ENABLE_RESTORE_VGPR_EXEC_HI:
+	s_mov_b32	exec_hi, 0xFFFFFFFF
+L_RESTORE_VGPR_NORMAL:
+	s_getreg_b32	s_restore_alloc_size, hwreg(HW_REG_GPR_ALLOC,SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SHIFT,SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SIZE)
+	s_add_u32	s_restore_alloc_size, s_restore_alloc_size, 1
+	s_lshl_b32	s_restore_alloc_size, s_restore_alloc_size, 2		//Number of VGPRs = (vgpr_size + 1) * 4    (non-zero value)
+	//determine it is wave32 or wave64
+	s_lshr_b32	m0, s_restore_size, S_WAVE_SIZE
+	s_and_b32	m0, m0, 1
+	s_cmp_eq_u32	m0, 1
+	s_cbranch_scc1	L_RESTORE_VGPR_WAVE64
+
+	s_mov_b32	s_restore_buf_rsrc2, 0x1000000				//NUM_RECORDS in bytes
+
+	// VGPR load using dw burst
+	s_mov_b32	s_restore_mem_offset_save, s_restore_mem_offset		// restore start with v1, v0 will be the last
+	s_add_u32	s_restore_mem_offset, s_restore_mem_offset, 128*4
+	s_mov_b32	m0, 4							//VGPR initial index value = 4
+
+L_RESTORE_VGPR_WAVE32_LOOP:
+	buffer_load_dword	v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset slc:1 glc:1
+	buffer_load_dword	v1, v0, s_restore_buf_rsrc0, s_restore_mem_offset slc:1 glc:1 offset:128
+	buffer_load_dword	v2, v0, s_restore_buf_rsrc0, s_restore_mem_offset slc:1 glc:1 offset:128*2
+	buffer_load_dword	v3, v0, s_restore_buf_rsrc0, s_restore_mem_offset slc:1 glc:1 offset:128*3
+	s_waitcnt	vmcnt(0)
+	v_movreld_b32	v0, v0							//v[0+m0] = v0
+	v_movreld_b32	v1, v1
+	v_movreld_b32	v2, v2
+	v_movreld_b32	v3, v3
+	s_add_u32	m0, m0, 4						//next vgpr index
+	s_add_u32	s_restore_mem_offset, s_restore_mem_offset, 128*4	//every buffer_load_dword does 128 bytes
+	s_cmp_lt_u32	m0, s_restore_alloc_size				//scc = (m0 < s_restore_alloc_size) ? 1 : 0
+	s_cbranch_scc1	L_RESTORE_VGPR_WAVE32_LOOP				//VGPR restore (except v0) is complete?
+
+	/* VGPR restore on v0 */
+	buffer_load_dword	v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save slc:1 glc:1
+	buffer_load_dword	v1, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save slc:1 glc:1 offset:128
+	buffer_load_dword	v2, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save slc:1 glc:1 offset:128*2
+	buffer_load_dword	v3, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save slc:1 glc:1 offset:128*3
+
+	s_branch	L_RESTORE_SGPR
+
+L_RESTORE_VGPR_WAVE64:
+	s_mov_b32	s_restore_buf_rsrc2, 0x1000000				//NUM_RECORDS in bytes
+
+	// VGPR load using dw burst
+	s_mov_b32	s_restore_mem_offset_save, s_restore_mem_offset		// restore start with v4, v0 will be the last
+	s_add_u32	s_restore_mem_offset, s_restore_mem_offset, 256*4
+	s_mov_b32	m0, 4							//VGPR initial index value = 4
+
+L_RESTORE_VGPR_WAVE64_LOOP:
+	buffer_load_dword	v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset slc:1 glc:1
+	buffer_load_dword	v1, v0, s_restore_buf_rsrc0, s_restore_mem_offset slc:1 glc:1 offset:256
+	buffer_load_dword	v2, v0, s_restore_buf_rsrc0, s_restore_mem_offset slc:1 glc:1 offset:256*2
+	buffer_load_dword	v3, v0, s_restore_buf_rsrc0, s_restore_mem_offset slc:1 glc:1 offset:256*3
+	s_waitcnt	vmcnt(0)
+	v_movreld_b32	v0, v0							//v[0+m0] = v0
+	v_movreld_b32	v1, v1
+	v_movreld_b32	v2, v2
+	v_movreld_b32	v3, v3
+	s_add_u32	m0, m0, 4						//next vgpr index
+	s_add_u32	s_restore_mem_offset, s_restore_mem_offset, 256*4	//every buffer_load_dword does 256 bytes
+	s_cmp_lt_u32	m0, s_restore_alloc_size				//scc = (m0 < s_restore_alloc_size) ? 1 : 0
+	s_cbranch_scc1	L_RESTORE_VGPR_WAVE64_LOOP				//VGPR restore (except v0) is complete?
+
+	//Below part will be the restore shared vgpr part (new for gfx10)
+	s_getreg_b32	s_restore_alloc_size, hwreg(HW_REG_LDS_ALLOC,SQ_WAVE_LDS_ALLOC_VGPR_SHARED_SIZE_SHIFT,SQ_WAVE_LDS_ALLOC_VGPR_SHARED_SIZE_SIZE)	//shared_vgpr_size
+	s_and_b32	s_restore_alloc_size, s_restore_alloc_size, 0xFFFFFFFF	//shared_vgpr_size is zero?
+	s_cbranch_scc0	L_RESTORE_V0						//no shared_vgpr used?
+	s_lshl_b32	s_restore_alloc_size, s_restore_alloc_size, 3		//Number of SHARED_VGPRs = shared_vgpr_size * 8    (non-zero value)
+	//m0 now has the value of normal vgpr count, just add the m0 with shared_vgpr count to get the total count.
+	//restore shared_vgpr will start from the index of m0
+	s_add_u32	s_restore_alloc_size, s_restore_alloc_size, m0
+	s_mov_b32	exec_lo, 0xFFFFFFFF
+	s_mov_b32	exec_hi, 0x00000000
+L_RESTORE_SHARED_VGPR_WAVE64_LOOP:
+	buffer_load_dword	v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset slc:1 glc:1
+	s_waitcnt	vmcnt(0)
+	v_movreld_b32	v0, v0							//v[0+m0] = v0
+	s_add_u32	m0, m0, 1						//next vgpr index
+	s_add_u32	s_restore_mem_offset, s_restore_mem_offset, 128
+	s_cmp_lt_u32	m0, s_restore_alloc_size				//scc = (m0 < s_restore_alloc_size) ? 1 : 0
+	s_cbranch_scc1	L_RESTORE_SHARED_VGPR_WAVE64_LOOP			//VGPR restore (except v0) is complete?
+
+	s_mov_b32	exec_hi, 0xFFFFFFFF					//restore back exec_hi before restoring V0!!
+
+	/* VGPR restore on v0 */
+L_RESTORE_V0:
+	buffer_load_dword	v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save slc:1 glc:1
+	buffer_load_dword	v1, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save slc:1 glc:1 offset:256
+	buffer_load_dword	v2, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save slc:1 glc:1 offset:256*2
+	buffer_load_dword	v3, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save slc:1 glc:1 offset:256*3
+
+	/* restore SGPRs */
+	//will be 2+8+16*6
+	// SGPR SR memory offset : size(VGPR)+size(SVGPR)
+L_RESTORE_SGPR:
+	get_vgpr_size_bytes(s_restore_mem_offset, s_restore_size)
+	get_svgpr_size_bytes(s_restore_tmp)
+	s_add_u32	s_restore_mem_offset, s_restore_mem_offset, s_restore_tmp
+	s_add_u32	s_restore_mem_offset, s_restore_mem_offset, get_sgpr_size_bytes()
+	s_sub_u32	s_restore_mem_offset, s_restore_mem_offset, 20*4	//s108~s127 is not saved
+
+	s_mov_b32	s_restore_buf_rsrc2, 0x1000000				//NUM_RECORDS in bytes
+
+	s_mov_b32	m0, s_sgpr_save_num
+
+	read_4sgpr_from_mem(s0, s_restore_buf_rsrc0, s_restore_mem_offset)
+	s_waitcnt	lgkmcnt(0)
+
+	s_sub_u32	m0, m0, 4						// Restore from S[0] to S[104]
+	s_nop		0							// hazard SALU M0=> S_MOVREL
+
+	s_movreld_b64	s0, s0							//s[0+m0] = s0
+	s_movreld_b64	s2, s2
+
+	read_8sgpr_from_mem(s0, s_restore_buf_rsrc0, s_restore_mem_offset)
+	s_waitcnt	lgkmcnt(0)
+
+	s_sub_u32	m0, m0, 8						// Restore from S[0] to S[96]
+	s_nop		0							// hazard SALU M0=> S_MOVREL
+
+	s_movreld_b64	s0, s0							//s[0+m0] = s0
+	s_movreld_b64	s2, s2
+	s_movreld_b64	s4, s4
+	s_movreld_b64	s6, s6
+
+ L_RESTORE_SGPR_LOOP:
+	read_16sgpr_from_mem(s0, s_restore_buf_rsrc0, s_restore_mem_offset)
+	s_waitcnt	lgkmcnt(0)
+
+	s_sub_u32	m0, m0, 16						// Restore from S[n] to S[0]
+	s_nop		0							// hazard SALU M0=> S_MOVREL
+
+	s_movreld_b64	s0, s0							//s[0+m0] = s0
+	s_movreld_b64	s2, s2
+	s_movreld_b64	s4, s4
+	s_movreld_b64	s6, s6
+	s_movreld_b64	s8, s8
+	s_movreld_b64	s10, s10
+	s_movreld_b64	s12, s12
+	s_movreld_b64	s14, s14
+
+	s_cmp_eq_u32	m0, 0							//scc = (m0 < s_sgpr_save_num) ? 1 : 0
+	s_cbranch_scc0	L_RESTORE_SGPR_LOOP
+
+	/* restore HW registers */
+L_RESTORE_HWREG:
+	// HWREG SR memory offset : size(VGPR)+size(SVGPR)+size(SGPR)
+	get_vgpr_size_bytes(s_restore_mem_offset, s_restore_size)
+	get_svgpr_size_bytes(s_restore_tmp)
+	s_add_u32	s_restore_mem_offset, s_restore_mem_offset, s_restore_tmp
+	s_add_u32	s_restore_mem_offset, s_restore_mem_offset, get_sgpr_size_bytes()
+
+	s_mov_b32	s_restore_buf_rsrc2, 0x1000000				//NUM_RECORDS in bytes
+
+	read_hwreg_from_mem(s_restore_m0, s_restore_buf_rsrc0, s_restore_mem_offset)
+	read_hwreg_from_mem(s_restore_pc_lo, s_restore_buf_rsrc0, s_restore_mem_offset)
+	read_hwreg_from_mem(s_restore_pc_hi, s_restore_buf_rsrc0, s_restore_mem_offset)
+	read_hwreg_from_mem(s_restore_exec_lo, s_restore_buf_rsrc0, s_restore_mem_offset)
+	read_hwreg_from_mem(s_restore_exec_hi, s_restore_buf_rsrc0, s_restore_mem_offset)
+	read_hwreg_from_mem(s_restore_status, s_restore_buf_rsrc0, s_restore_mem_offset)
+	read_hwreg_from_mem(s_restore_trapsts, s_restore_buf_rsrc0, s_restore_mem_offset)
+	read_hwreg_from_mem(s_restore_xnack_mask, s_restore_buf_rsrc0, s_restore_mem_offset)
+	read_hwreg_from_mem(s_restore_mode, s_restore_buf_rsrc0, s_restore_mem_offset)
+	read_hwreg_from_mem(s_restore_flat_scratch, s_restore_buf_rsrc0, s_restore_mem_offset)
+	s_waitcnt	lgkmcnt(0)
+
+	s_setreg_b32	hwreg(HW_REG_SHADER_FLAT_SCRATCH_LO), s_restore_flat_scratch
+
+	read_hwreg_from_mem(s_restore_flat_scratch, s_restore_buf_rsrc0, s_restore_mem_offset)
+	s_waitcnt	lgkmcnt(0)						//from now on, it is safe to restore STATUS and IB_STS
+
+	s_setreg_b32	hwreg(HW_REG_SHADER_FLAT_SCRATCH_HI), s_restore_flat_scratch
+
+	s_mov_b32	s_restore_tmp, s_restore_pc_hi
+	s_and_b32	s_restore_pc_hi, s_restore_tmp, 0x0000ffff		//pc[47:32] //Do it here in order not to affect STATUS
+
+	s_mov_b32	m0, s_restore_m0
+	s_mov_b32	exec_lo, s_restore_exec_lo
+	s_mov_b32	exec_hi, s_restore_exec_hi
+
+	s_and_b32	s_restore_m0, SQ_WAVE_TRAPSTS_PRE_SAVECTX_MASK, s_restore_trapsts
 	s_setreg_b32	hwreg(HW_REG_TRAPSTS, SQ_WAVE_TRAPSTS_PRE_SAVECTX_SHIFT, SQ_WAVE_TRAPSTS_PRE_SAVECTX_SIZE), s_restore_m0
-    s_setreg_b32    hwreg(HW_REG_SHADER_XNACK_MASK), s_restore_xnack_mask         //restore xnack_mask
-	s_and_b32		s_restore_m0, SQ_WAVE_TRAPSTS_POST_SAVECTX_MASK, s_restore_trapsts
-	s_lshr_b32		s_restore_m0, s_restore_m0, SQ_WAVE_TRAPSTS_POST_SAVECTX_SHIFT
+	s_setreg_b32	hwreg(HW_REG_SHADER_XNACK_MASK), s_restore_xnack_mask
+	s_and_b32	s_restore_m0, SQ_WAVE_TRAPSTS_POST_SAVECTX_MASK, s_restore_trapsts
+	s_lshr_b32	s_restore_m0, s_restore_m0, SQ_WAVE_TRAPSTS_POST_SAVECTX_SHIFT
 	s_setreg_b32	hwreg(HW_REG_TRAPSTS, SQ_WAVE_TRAPSTS_POST_SAVECTX_SHIFT, SQ_WAVE_TRAPSTS_POST_SAVECTX_SIZE), s_restore_m0
-	//s_setreg_b32 	hwreg(HW_REG_TRAPSTS), 	s_restore_trapsts      //don't overwrite SAVECTX bit as it may be set through external SAVECTX during restore
-	s_setreg_b32 	hwreg(HW_REG_MODE), 	s_restore_mode
-	//reuse s_restore_m0 as a temp register
-	s_and_b32		s_restore_m0, s_restore_pc_hi, S_SAVE_PC_HI_RCNT_MASK
-	s_lshr_b32		s_restore_m0, s_restore_m0, S_SAVE_PC_HI_RCNT_SHIFT
-	s_lshl_b32		s_restore_m0, s_restore_m0, SQ_WAVE_IB_STS_RCNT_SHIFT
-	s_mov_b32		s_restore_tmp, 0x0																				//IB_STS is zero
-	s_or_b32		s_restore_tmp, s_restore_tmp, s_restore_m0
-	s_and_b32		s_restore_m0, s_restore_pc_hi, S_SAVE_PC_HI_FIRST_REPLAY_MASK
-	s_lshr_b32		s_restore_m0, s_restore_m0, S_SAVE_PC_HI_FIRST_REPLAY_SHIFT
-	s_lshl_b32		s_restore_m0, s_restore_m0, SQ_WAVE_IB_STS_FIRST_REPLAY_SHIFT
-	s_or_b32		s_restore_tmp, s_restore_tmp, s_restore_m0
-    s_and_b32       s_restore_m0, s_restore_status, SQ_WAVE_STATUS_INST_ATC_MASK 
-    s_lshr_b32		s_restore_m0, s_restore_m0, SQ_WAVE_STATUS_INST_ATC_SHIFT
-	s_setreg_b32 	hwreg(HW_REG_IB_STS), 	s_restore_tmp
-	s_setreg_b32 	hwreg(HW_REG_STATUS), 	s_restore_status
-
-	s_barrier													//barrier to ensure the readiness of LDS before access attemps from any other wave in the same TG //FIXME not performance-optimal at this time
-	
-	
-//	s_rfe_b64 s_restore_pc_lo                              		//Return to the main shader program and resume execution
-    s_rfe_b64  s_restore_pc_lo            // s_restore_m0[0] is used to set STATUS.inst_atc 
-
-
-/**************************************************************************/
-/*                     	the END								              */
-/**************************************************************************/	
-L_END_PGM:	
+	s_setreg_b32	hwreg(HW_REG_MODE), s_restore_mode
+	s_and_b32	s_restore_m0, s_restore_tmp, S_SAVE_PC_HI_RCNT_MASK
+	s_lshr_b32	s_restore_m0, s_restore_m0, S_SAVE_PC_HI_RCNT_SHIFT
+	s_lshl_b32	s_restore_m0, s_restore_m0, SQ_WAVE_IB_STS_RCNT_SHIFT
+	s_mov_b32	s_restore_mode, 0x0
+	s_or_b32	s_restore_mode, s_restore_mode, s_restore_m0
+	s_and_b32	s_restore_m0, s_restore_tmp, S_SAVE_PC_HI_FIRST_REPLAY_MASK
+	s_lshr_b32	s_restore_m0, s_restore_m0, S_SAVE_PC_HI_FIRST_REPLAY_SHIFT
+	s_lshl_b32	s_restore_m0, s_restore_m0, SQ_WAVE_IB_STS_FIRST_REPLAY_SHIFT
+	s_or_b32	s_restore_mode, s_restore_mode, s_restore_m0
+	s_and_b32	s_restore_m0, s_restore_tmp, S_SAVE_PC_HI_REPLAY_W64H_MASK
+	s_lshr_b32	s_restore_m0, s_restore_m0, S_SAVE_PC_HI_REPLAY_W64H_SHIFT
+	s_lshl_b32	s_restore_m0, s_restore_m0, SQ_WAVE_IB_STS_REPLAY_W64H_SHIFT
+	s_or_b32	s_restore_mode, s_restore_mode, s_restore_m0
+
+	s_and_b32	s_restore_m0, s_restore_status, SQ_WAVE_STATUS_INST_ATC_MASK
+	s_lshr_b32	s_restore_m0, s_restore_m0, SQ_WAVE_STATUS_INST_ATC_SHIFT
+	s_setreg_b32 	hwreg(HW_REG_IB_STS), s_restore_mode
+
+	s_and_b64	exec, exec, exec					// Restore STATUS.EXECZ, not writable by s_setreg_b32
+	s_and_b64	vcc, vcc, vcc						// Restore STATUS.VCCZ, not writable by s_setreg_b32
+	s_setreg_b32	hwreg(HW_REG_STATUS), s_restore_status			// SCC is included, which is changed by previous salu
+
+	s_barrier								//barrier to ensure the readiness of LDS before access attemps from any other wave in the same TG
+
+	s_rfe_b64	s_restore_pc_lo						//Return to the main shader program and resume execution
+
+L_END_PGM:
 	s_endpgm
-	
-end	
-
-
-/**************************************************************************/
-/*                     	the helper functions							  */
-/**************************************************************************/
-function write_sgpr_to_mem_wave32(s, s_rsrc, s_mem_offset, use_sqc, use_mtbuf)
-	if (use_sqc)
-		s_mov_b32 exec_lo, m0					//assuming exec_lo is not needed anymore from this point on
-		s_mov_b32 m0, s_mem_offset
-		s_buffer_store_dword s, s_rsrc, m0		glc:1	
-		s_add_u32		s_mem_offset, s_mem_offset, 4
-		s_mov_b32	m0, exec_lo
-    elsif (use_mtbuf)
-        v_mov_b32	v0,	s
-        tbuffer_store_format_x v0, v0, s_rsrc, s_mem_offset format:BUF_NUM_FORMAT_FLOAT format: BUF_DATA_FORMAT_32 slc:1 glc:1
-		s_add_u32		s_mem_offset, s_mem_offset, 128
-    else 
-        v_mov_b32	v0,	s
-		buffer_store_dword	v0, v0, s_rsrc, s_mem_offset	slc:1 glc:1
-        s_add_u32		s_mem_offset, s_mem_offset, 128
-	end
 end
 
-function write_sgpr_to_mem_wave64(s, s_rsrc, s_mem_offset, use_sqc, use_mtbuf)
-	if (use_sqc)
-		s_mov_b32 exec_lo, m0					//assuming exec_lo is not needed anymore from this point on
-		s_mov_b32 m0, s_mem_offset
-		s_buffer_store_dword s, s_rsrc, m0		glc:1	
-		s_add_u32		s_mem_offset, s_mem_offset, 4
-		s_mov_b32	m0, exec_lo
-    elsif (use_mtbuf)
-        v_mov_b32	v0,	s
-        tbuffer_store_format_x v0, v0, s_rsrc, s_mem_offset format:BUF_NUM_FORMAT_FLOAT format: BUF_DATA_FORMAT_32 slc:1 glc:1
-		s_add_u32		s_mem_offset, s_mem_offset, 256
-    else 
-        v_mov_b32	v0,	s
-		buffer_store_dword	v0, v0, s_rsrc, s_mem_offset	slc:1 glc:1
-        s_add_u32		s_mem_offset, s_mem_offset, 256
-	end
+function write_hwreg_to_mem(s, s_rsrc, s_mem_offset)
+	s_mov_b32	exec_lo, m0
+	s_mov_b32	m0, s_mem_offset
+	s_buffer_store_dword	s, s_rsrc, m0 glc:1
+	s_add_u32	s_mem_offset, s_mem_offset, 4
+	s_mov_b32	m0, exec_lo
+end
+
+
+function write_16sgpr_to_mem(s, s_rsrc, s_mem_offset)
+	s_buffer_store_dwordx4	s[0], s_rsrc, 0 glc:1
+	s_buffer_store_dwordx4	s[4], s_rsrc, 16 glc:1
+	s_buffer_store_dwordx4	s[8], s_rsrc, 32 glc:1
+	s_buffer_store_dwordx4	s[12], s_rsrc, 48 glc:1
+	s_add_u32	s_rsrc[0], s_rsrc[0], 4*16
+	s_addc_u32	s_rsrc[1], s_rsrc[1], 0x0
+end
+
+function write_12sgpr_to_mem(s, s_rsrc, s_mem_offset)
+	s_buffer_store_dwordx4	s[0], s_rsrc, 0 glc:1
+	s_buffer_store_dwordx4	s[4], s_rsrc, 16 glc:1
+	s_buffer_store_dwordx4	s[8], s_rsrc, 32 glc:1
+	s_add_u32	s_rsrc[0], s_rsrc[0], 4*12
+	s_addc_u32	s_rsrc[1], s_rsrc[1], 0x0
+end
+
+
+function read_hwreg_from_mem(s, s_rsrc, s_mem_offset)
+	s_buffer_load_dword	s, s_rsrc, s_mem_offset glc:1
+	s_add_u32	s_mem_offset, s_mem_offset, 4
 end
 
-function read_sgpr_from_mem_wave32(s, s_rsrc, s_mem_offset, use_sqc)
-	s_buffer_load_dword s, s_rsrc, s_mem_offset		glc:1
-	if (use_sqc)
-		s_add_u32		s_mem_offset, s_mem_offset, 4
-	else
-        s_add_u32		s_mem_offset, s_mem_offset, 128
-	end
+function read_16sgpr_from_mem(s, s_rsrc, s_mem_offset)
+	s_sub_u32	s_mem_offset, s_mem_offset, 4*16
+	s_buffer_load_dwordx16	s, s_rsrc, s_mem_offset glc:1
 end
 
-function read_sgpr_from_mem_wave64(s, s_rsrc, s_mem_offset, use_sqc)
-	s_buffer_load_dword s, s_rsrc, s_mem_offset		glc:1
-	if (use_sqc)
-		s_add_u32		s_mem_offset, s_mem_offset, 4
-	else
-        s_add_u32		s_mem_offset, s_mem_offset, 256
-	end
+function read_8sgpr_from_mem(s, s_rsrc, s_mem_offset)
+	s_sub_u32	s_mem_offset, s_mem_offset, 4*8
+	s_buffer_load_dwordx8	s, s_rsrc, s_mem_offset glc:1
 end
 
+function read_4sgpr_from_mem(s, s_rsrc, s_mem_offset)
+	s_sub_u32	s_mem_offset, s_mem_offset, 4*4
+	s_buffer_load_dwordx4	s, s_rsrc, s_mem_offset glc:1
+end
+
+
+function get_lds_size_bytes(s_lds_size_byte)
+	s_getreg_b32	s_lds_size_byte, hwreg(HW_REG_LDS_ALLOC, SQ_WAVE_LDS_ALLOC_LDS_SIZE_SHIFT, SQ_WAVE_LDS_ALLOC_LDS_SIZE_SIZE)
+	s_lshl_b32	s_lds_size_byte, s_lds_size_byte, 8			//LDS size in dwords = lds_size * 64 *4Bytes // granularity 64DW
+end
+
+function get_vgpr_size_bytes(s_vgpr_size_byte, s_size)
+	s_getreg_b32	s_vgpr_size_byte, hwreg(HW_REG_GPR_ALLOC,SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SHIFT,SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SIZE)
+	s_add_u32	s_vgpr_size_byte, s_vgpr_size_byte, 1
+	s_lshr_b32	m0, s_size, S_WAVE_SIZE
+	s_and_b32	m0, m0, 1
+	s_cmp_eq_u32	m0, 1
+	s_cbranch_scc1	L_ENABLE_SHIFT_W64
+	s_lshl_b32	s_vgpr_size_byte, s_vgpr_size_byte, (2+7)		//Number of VGPRs = (vgpr_size + 1) * 4 * 32 * 4   (non-zero value)
+	s_branch	L_SHIFT_DONE
+L_ENABLE_SHIFT_W64:
+	s_lshl_b32	s_vgpr_size_byte, s_vgpr_size_byte, (2+8)		//Number of VGPRs = (vgpr_size + 1) * 4 * 64 * 4   (non-zero value)
+L_SHIFT_DONE:
+end
+
+function get_svgpr_size_bytes(s_svgpr_size_byte)
+	s_getreg_b32	s_svgpr_size_byte, hwreg(HW_REG_LDS_ALLOC,SQ_WAVE_LDS_ALLOC_VGPR_SHARED_SIZE_SHIFT,SQ_WAVE_LDS_ALLOC_VGPR_SHARED_SIZE_SIZE)
+	s_lshl_b32	s_svgpr_size_byte, s_svgpr_size_byte, (3+7)
+end
+
+function get_sgpr_size_bytes
+	return 512
+end
+
+function get_hwreg_size_bytes
+	return 128
+end
diff --git a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx8.asm b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx8.asm
index a47f5b933120..b195b7cd8a17 100644
--- a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx8.asm
+++ b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx8.asm
@@ -24,78 +24,6 @@
  * PROJECT=vi ./sp3 cwsr_trap_handler_gfx8.asm -hex tmp.hex
  */
 
-/* HW (VI) source code for CWSR trap handler */
-/* Version 18 + multiple trap handler */
-
-// this performance-optimal version was originally from Seven Xu at SRDC
-
-// Revison #18   --...
-/* Rev History
-** #1. Branch from gc dv.   //gfxip/gfx8/main/src/test/suites/block/cs/sr/cs_trap_handler.sp3#1,#50, #51, #52-53(Skip, Already Fixed by PV), #54-56(merged),#57-58(mergerd, skiped-already fixed by PV)
-** #4. SR Memory Layout:
-**             1. VGPR-SGPR-HWREG-{LDS}
-**             2. tba_hi.bits.26 - reconfigured as the first wave in tg bits, for defer Save LDS for a threadgroup.. performance concern..
-** #5. Update: 1. Accurate g8sr_ts_save_d timestamp
-** #6. Update: 1. Fix s_barrier usage; 2. VGPR s/r using swizzle buffer?(NoNeed, already matched the swizzle pattern, more investigation)
-** #7. Update: 1. don't barrier if noLDS
-** #8. Branch: 1. Branch to ver#0, which is very similar to gc dv version
-**             2. Fix SQ issue by s_sleep 2
-** #9. Update: 1. Fix scc restore failed issue, restore wave_status at last
-**             2. optimize s_buffer save by burst 16sgprs...
-** #10. Update 1. Optimize restore sgpr by busrt 16 sgprs.
-** #11. Update 1. Add 2 more timestamp for debug version
-** #12. Update 1. Add VGPR SR using DWx4, some case improve and some case drop performance
-** #13. Integ  1. Always use MUBUF for PV trap shader...
-** #14. Update 1. s_buffer_store soft clause...
-** #15. Update 1. PERF - sclar write with glc:0/mtype0 to allow L2 combine. perf improvement a lot.
-** #16. Update 1. PRRF - UNROLL LDS_DMA got 2500cycle save in IP tree
-** #17. Update 1. FUNC - LDS_DMA has issues while ATC, replace with ds_read/buffer_store for save part[TODO restore part]
-**             2. PERF - Save LDS before save VGPR to cover LDS save long latency...
-** #18. Update 1. FUNC - Implicitly estore STATUS.VCCZ, which is not writable by s_setreg_b32
-**             2. FUNC - Handle non-CWSR traps
-*/
-
-var G8SR_WDMEM_HWREG_OFFSET = 0
-var G8SR_WDMEM_SGPR_OFFSET  = 128  // in bytes
-
-// Keep definition same as the app shader, These 2 time stamps are part of the app shader... Should before any Save and after restore.
-
-var G8SR_DEBUG_TIMESTAMP = 0
-var G8SR_DEBUG_TS_SAVE_D_OFFSET = 40*4  // ts_save_d timestamp offset relative to SGPR_SR_memory_offset
-var s_g8sr_ts_save_s    = s[34:35]   // save start
-var s_g8sr_ts_sq_save_msg  = s[36:37]   // The save shader send SAVEWAVE msg to spi
-var s_g8sr_ts_spi_wrexec   = s[38:39]   // the SPI write the sr address to SQ
-var s_g8sr_ts_save_d    = s[40:41]   // save end
-var s_g8sr_ts_restore_s = s[42:43]   // restore start
-var s_g8sr_ts_restore_d = s[44:45]   // restore end
-
-var G8SR_VGPR_SR_IN_DWX4 = 0
-var G8SR_SAVE_BUF_RSRC_WORD1_STRIDE_DWx4 = 0x00100000    // DWx4 stride is 4*4Bytes
-var G8SR_RESTORE_BUF_RSRC_WORD1_STRIDE_DWx4  = G8SR_SAVE_BUF_RSRC_WORD1_STRIDE_DWx4
-
-
-/*************************************************************************/
-/*                  control on how to run the shader                     */
-/*************************************************************************/
-//any hack that needs to be made to run this code in EMU (either because various EMU code are not ready or no compute save & restore in EMU run)
-var EMU_RUN_HACK                    =   0
-var EMU_RUN_HACK_RESTORE_NORMAL     =   0
-var EMU_RUN_HACK_SAVE_NORMAL_EXIT   =   0
-var EMU_RUN_HACK_SAVE_SINGLE_WAVE   =   0
-var EMU_RUN_HACK_SAVE_FIRST_TIME    =   0                   //for interrupted restore in which the first save is through EMU_RUN_HACK
-var EMU_RUN_HACK_SAVE_FIRST_TIME_TBA_LO =   0                   //for interrupted restore in which the first save is through EMU_RUN_HACK
-var EMU_RUN_HACK_SAVE_FIRST_TIME_TBA_HI =   0                   //for interrupted restore in which the first save is through EMU_RUN_HACK
-var SAVE_LDS                        =   1
-var WG_BASE_ADDR_LO                 =   0x9000a000
-var WG_BASE_ADDR_HI                 =   0x0
-var WAVE_SPACE                      =   0x5000              //memory size that each wave occupies in workgroup state mem
-var CTX_SAVE_CONTROL                =   0x0
-var CTX_RESTORE_CONTROL             =   CTX_SAVE_CONTROL
-var SIM_RUN_HACK                    =   0                   //any hack that needs to be made to run this code in SIM (either because various RTL code are not ready or no compute save & restore in RTL run)
-var SGPR_SAVE_USE_SQC               =   1                   //use SQC D$ to do the write
-var USE_MTBUF_INSTEAD_OF_MUBUF      =   0                   //because TC EMU currently asserts on 0 of // overload DFMT field to carry 4 more bits of stride for MUBUF opcodes
-var SWIZZLE_EN                      =   0                   //whether we use swizzled buffer addressing
-
 /**************************************************************************/
 /*                      variables                                         */
 /**************************************************************************/
@@ -226,16 +154,7 @@ shader main
   type(CS)
 
 
-    if ((EMU_RUN_HACK) && (!EMU_RUN_HACK_RESTORE_NORMAL))                   //hack to use trap_id for determining save/restore
-        //FIXME VCCZ un-init assertion s_getreg_b32     s_save_status, hwreg(HW_REG_STATUS)         //save STATUS since we will change SCC
-        s_and_b32 s_save_tmp, s_save_pc_hi, 0xffff0000              //change SCC
-        s_cmp_eq_u32 s_save_tmp, 0x007e0000                         //Save: trap_id = 0x7e. Restore: trap_id = 0x7f.
-        s_cbranch_scc0 L_JUMP_TO_RESTORE                            //do not need to recover STATUS here  since we are going to RESTORE
-        //FIXME  s_setreg_b32   hwreg(HW_REG_STATUS),   s_save_status       //need to recover STATUS since we are going to SAVE
-        s_branch L_SKIP_RESTORE                                     //NOT restore, SAVE actually
-    else
         s_branch L_SKIP_RESTORE                                     //NOT restore. might be a regular trap or save
-    end
 
 L_JUMP_TO_RESTORE:
     s_branch L_RESTORE                                              //restore
@@ -249,7 +168,7 @@ L_SKIP_RESTORE:
     s_cbranch_scc1  L_SAVE                                      //this is the operation for save
 
     // *********    Handle non-CWSR traps       *******************
-if (!EMU_RUN_HACK)
+
     /* read tba and tma for next level trap handler, ttmp4 is used as s_save_status */
     s_load_dwordx4  [ttmp8,ttmp9,ttmp10, ttmp11], [tma_lo,tma_hi], 0
     s_waitcnt lgkmcnt(0)
@@ -268,7 +187,7 @@ L_EXCP_CASE:
     s_and_b32   ttmp1, ttmp1, 0xFFFF
     set_status_without_spi_prio(s_save_status, ttmp2) //restore HW status(SCC)
     s_rfe_b64       [ttmp0, ttmp1]
-end
+
     // *********        End handling of non-CWSR traps   *******************
 
 /**************************************************************************/
@@ -276,12 +195,6 @@ end
 /**************************************************************************/
 
 L_SAVE:
-
-if G8SR_DEBUG_TIMESTAMP
-        s_memrealtime   s_g8sr_ts_save_s
-        s_waitcnt lgkmcnt(0)         //FIXME, will cause xnack??
-end
-
     s_mov_b32       s_save_tmp, 0                                                           //clear saveCtx bit
     s_setreg_b32    hwreg(HW_REG_TRAPSTS, SQ_WAVE_TRAPSTS_SAVECTX_SHIFT, 1), s_save_tmp     //clear saveCtx bit
 
@@ -303,16 +216,7 @@ end
     s_mov_b32       s_save_exec_hi, exec_hi
     s_mov_b64       exec,   0x0                                                             //clear EXEC to get ready to receive
 
-if G8SR_DEBUG_TIMESTAMP
-        s_memrealtime  s_g8sr_ts_sq_save_msg
-        s_waitcnt lgkmcnt(0)
-end
-
-    if (EMU_RUN_HACK)
-
-    else
         s_sendmsg   sendmsg(MSG_SAVEWAVE)  //send SPI a message and wait for SPI's write to EXEC
-    end
 
     // Set SPI_PRIO=2 to avoid starving instruction fetch in the waves we're waiting for.
     s_or_b32 s_save_tmp, s_save_status, (2 << SQ_WAVE_STATUS_SPI_PRIO_SHIFT)
@@ -321,36 +225,9 @@ end
   L_SLEEP:
     s_sleep 0x2                // sleep 1 (64clk) is not enough for 8 waves per SIMD, which will cause SQ hang, since the 7,8th wave could not get arbit to exec inst, while other waves are stuck into the sleep-loop and waiting for wrexec!=0
 
-    if (EMU_RUN_HACK)
-
-    else
         s_cbranch_execz L_SLEEP
-    end
-
-if G8SR_DEBUG_TIMESTAMP
-        s_memrealtime  s_g8sr_ts_spi_wrexec
-        s_waitcnt lgkmcnt(0)
-end
 
     /*      setup Resource Contants    */
-    if ((EMU_RUN_HACK) && (!EMU_RUN_HACK_SAVE_SINGLE_WAVE))
-        //calculate wd_addr using absolute thread id
-        v_readlane_b32 s_save_tmp, v9, 0
-        s_lshr_b32 s_save_tmp, s_save_tmp, 6
-        s_mul_i32 s_save_tmp, s_save_tmp, WAVE_SPACE
-        s_add_i32 s_save_spi_init_lo, s_save_tmp, WG_BASE_ADDR_LO
-        s_mov_b32 s_save_spi_init_hi, WG_BASE_ADDR_HI
-        s_and_b32 s_save_spi_init_hi, s_save_spi_init_hi, CTX_SAVE_CONTROL
-    else
-    end
-    if ((EMU_RUN_HACK) && (EMU_RUN_HACK_SAVE_SINGLE_WAVE))
-        s_add_i32 s_save_spi_init_lo, s_save_tmp, WG_BASE_ADDR_LO
-        s_mov_b32 s_save_spi_init_hi, WG_BASE_ADDR_HI
-        s_and_b32 s_save_spi_init_hi, s_save_spi_init_hi, CTX_SAVE_CONTROL
-    else
-    end
-
-
     s_mov_b32       s_save_buf_rsrc0,   s_save_spi_init_lo                                                      //base_addr_lo
     s_and_b32       s_save_buf_rsrc1,   s_save_spi_init_hi, 0x0000FFFF                                          //base_addr_hi
     s_or_b32        s_save_buf_rsrc1,   s_save_buf_rsrc1,  S_SAVE_BUF_RSRC_WORD1_STRIDE
@@ -383,22 +260,10 @@ end
 
 
     s_mov_b32       s_save_buf_rsrc2, 0x4                               //NUM_RECORDS   in bytes
-    if (SWIZZLE_EN)
-        s_add_u32       s_save_buf_rsrc2, s_save_buf_rsrc2, 0x0                     //FIXME need to use swizzle to enable bounds checking?
-    else
         s_mov_b32       s_save_buf_rsrc2,  0x1000000                                //NUM_RECORDS in bytes
-    end
 
 
     write_hwreg_to_mem(s_save_m0, s_save_buf_rsrc0, s_save_mem_offset)                  //M0
-
-    if ((EMU_RUN_HACK) && (EMU_RUN_HACK_SAVE_FIRST_TIME))
-        s_add_u32 s_save_pc_lo, s_save_pc_lo, 4             //pc[31:0]+4
-        s_addc_u32 s_save_pc_hi, s_save_pc_hi, 0x0          //carry bit over
-        s_mov_b32   tba_lo, EMU_RUN_HACK_SAVE_FIRST_TIME_TBA_LO
-        s_mov_b32   tba_hi, EMU_RUN_HACK_SAVE_FIRST_TIME_TBA_HI
-    end
-
     write_hwreg_to_mem(s_save_pc_lo, s_save_buf_rsrc0, s_save_mem_offset)                   //PC
     write_hwreg_to_mem(s_save_pc_hi, s_save_buf_rsrc0, s_save_mem_offset)
     write_hwreg_to_mem(s_save_exec_lo, s_save_buf_rsrc0, s_save_mem_offset)             //EXEC
@@ -440,18 +305,8 @@ end
     s_add_u32       s_save_alloc_size, s_save_alloc_size, 1
     s_lshl_b32      s_save_alloc_size, s_save_alloc_size, 4                         //Number of SGPRs = (sgpr_size + 1) * 16   (non-zero value)
 
-    if (SGPR_SAVE_USE_SQC)
         s_lshl_b32      s_save_buf_rsrc2,   s_save_alloc_size, 2                    //NUM_RECORDS in bytes
-    else
-        s_lshl_b32      s_save_buf_rsrc2,   s_save_alloc_size, 8                    //NUM_RECORDS in bytes (64 threads)
-    end
-
-    if (SWIZZLE_EN)
-        s_add_u32       s_save_buf_rsrc2, s_save_buf_rsrc2, 0x0                     //FIXME need to use swizzle to enable bounds checking?
-    else
         s_mov_b32       s_save_buf_rsrc2,  0x1000000                                //NUM_RECORDS in bytes
-    end
-
 
     // backup s_save_buf_rsrc0,1 to s_save_pc_lo/hi, since write_16sgpr_to_mem function will change the rsrc0
     //s_mov_b64 s_save_pc_lo, s_save_buf_rsrc0
@@ -490,30 +345,14 @@ end
     s_mov_b32       exec_lo, 0xFFFFFFFF                                             //need every thread from now on
     s_mov_b32       exec_hi, 0xFFFFFFFF
 
-    if (SWIZZLE_EN)
-        s_add_u32       s_save_buf_rsrc2, s_save_buf_rsrc2, 0x0                     //FIXME need to use swizzle to enable bounds checking?
-    else
         s_mov_b32       s_save_buf_rsrc2,  0x1000000                                //NUM_RECORDS in bytes
-    end
-
 
     // VGPR Allocated in 4-GPR granularity
 
-if G8SR_VGPR_SR_IN_DWX4
-        // the const stride for DWx4 is 4*4 bytes
-        s_and_b32 s_save_buf_rsrc1, s_save_buf_rsrc1, 0x0000FFFF   // reset const stride to 0
-        s_or_b32  s_save_buf_rsrc1, s_save_buf_rsrc1, G8SR_SAVE_BUF_RSRC_WORD1_STRIDE_DWx4  // const stride to 4*4 bytes
-
-        buffer_store_dwordx4 v0, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1
-
-        s_and_b32 s_save_buf_rsrc1, s_save_buf_rsrc1, 0x0000FFFF   // reset const stride to 0
-        s_or_b32  s_save_buf_rsrc1, s_save_buf_rsrc1, S_SAVE_BUF_RSRC_WORD1_STRIDE  // reset const stride to 4 bytes
-else
         buffer_store_dword v0, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1
         buffer_store_dword v1, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1  offset:256
         buffer_store_dword v2, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1  offset:256*2
         buffer_store_dword v3, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1  offset:256*3
-end
 
 
 
@@ -549,64 +388,10 @@ end
     s_add_u32 s_save_mem_offset, s_save_mem_offset, get_hwreg_size_bytes()
 
 
-    if (SWIZZLE_EN)
-        s_add_u32       s_save_buf_rsrc2, s_save_buf_rsrc2, 0x0       //FIXME need to use swizzle to enable bounds checking?
-    else
         s_mov_b32       s_save_buf_rsrc2,  0x1000000                  //NUM_RECORDS in bytes
-    end
-
     s_mov_b32       m0, 0x0                                               //lds_offset initial value = 0
 
 
-var LDS_DMA_ENABLE = 0
-var UNROLL = 0
-if UNROLL==0 && LDS_DMA_ENABLE==1
-        s_mov_b32  s3, 256*2
-        s_nop 0
-        s_nop 0
-        s_nop 0
-  L_SAVE_LDS_LOOP:
-        //TODO: looks the 2 buffer_store/load clause for s/r will hurt performance.???
-    if (SAVE_LDS)     //SPI always alloc LDS space in 128DW granularity
-            buffer_store_lds_dword s_save_buf_rsrc0, s_save_mem_offset lds:1            // first 64DW
-            buffer_store_lds_dword s_save_buf_rsrc0, s_save_mem_offset lds:1 offset:256 // second 64DW
-    end
-
-    s_add_u32       m0, m0, s3                                          //every buffer_store_lds does 256 bytes
-    s_add_u32       s_save_mem_offset, s_save_mem_offset, s3                            //mem offset increased by 256 bytes
-    s_cmp_lt_u32    m0, s_save_alloc_size                                               //scc=(m0 < s_save_alloc_size) ? 1 : 0
-    s_cbranch_scc1  L_SAVE_LDS_LOOP                                                     //LDS save is complete?
-
-elsif LDS_DMA_ENABLE==1 && UNROLL==1 // UNROOL  , has ichace miss
-      // store from higest LDS address to lowest
-      s_mov_b32  s3, 256*2
-      s_sub_u32  m0, s_save_alloc_size, s3
-      s_add_u32 s_save_mem_offset, s_save_mem_offset, m0
-      s_lshr_b32 s_save_alloc_size, s_save_alloc_size, 9   // how many 128 trunks...
-      s_sub_u32 s_save_alloc_size, 128, s_save_alloc_size   // store from higheset addr to lowest
-      s_mul_i32 s_save_alloc_size, s_save_alloc_size, 6*4   // PC offset increment,  each LDS save block cost 6*4 Bytes instruction
-      s_add_u32 s_save_alloc_size, s_save_alloc_size, 3*4   //2is the below 2 inst...//s_addc and s_setpc
-      s_nop 0
-      s_nop 0
-      s_nop 0   //pad 3 dw to let LDS_DMA align with 64Bytes
-      s_getpc_b64 s[0:1]                              // reuse s[0:1], since s[0:1] already saved
-      s_add_u32   s0, s0,s_save_alloc_size
-      s_addc_u32  s1, s1, 0
-      s_setpc_b64 s[0:1]
-
-
-       for var i =0; i< 128; i++
-            // be careful to make here a 64Byte aligned address, which could improve performance...
-            buffer_store_lds_dword s_save_buf_rsrc0, s_save_mem_offset lds:1 offset:0           // first 64DW
-            buffer_store_lds_dword s_save_buf_rsrc0, s_save_mem_offset lds:1 offset:256           // second 64DW
-
-        if i!=127
-        s_sub_u32  m0, m0, s3      // use a sgpr to shrink 2DW-inst to 1DW inst to improve performance , i.e.  pack more LDS_DMA inst to one Cacheline
-            s_sub_u32  s_save_mem_offset, s_save_mem_offset,  s3
-            end
-       end
-
-else   // BUFFER_STORE
       v_mbcnt_lo_u32_b32 v2, 0xffffffff, 0x0
       v_mbcnt_hi_u32_b32 v3, 0xffffffff, v2     // tid
       v_mul_i32_i24 v2, v3, 8   // tid*8
@@ -628,8 +413,6 @@ L_SAVE_LDS_LOOP_VECTOR:
       // restore rsrc3
       s_mov_b32 s_save_buf_rsrc3, s0
 
-end
-
 L_SAVE_LDS_DONE:
 
 
@@ -647,44 +430,8 @@ L_SAVE_LDS_DONE:
     s_add_u32       s_save_alloc_size, s_save_alloc_size, 1
     s_lshl_b32      s_save_alloc_size, s_save_alloc_size, 2                         //Number of VGPRs = (vgpr_size + 1) * 4    (non-zero value)   //FIXME for GFX, zero is possible
     s_lshl_b32      s_save_buf_rsrc2,  s_save_alloc_size, 8                         //NUM_RECORDS in bytes (64 threads*4)
-    if (SWIZZLE_EN)
-        s_add_u32       s_save_buf_rsrc2, s_save_buf_rsrc2, 0x0                     //FIXME need to use swizzle to enable bounds checking?
-    else
         s_mov_b32       s_save_buf_rsrc2,  0x1000000                                //NUM_RECORDS in bytes
-    end
-
-
-    // VGPR Allocated in 4-GPR granularity
-
-if G8SR_VGPR_SR_IN_DWX4
-        // the const stride for DWx4 is 4*4 bytes
-        s_and_b32 s_save_buf_rsrc1, s_save_buf_rsrc1, 0x0000FFFF   // reset const stride to 0
-        s_or_b32  s_save_buf_rsrc1, s_save_buf_rsrc1, G8SR_SAVE_BUF_RSRC_WORD1_STRIDE_DWx4  // const stride to 4*4 bytes
-
-        s_mov_b32         m0, 4     // skip first 4 VGPRs
-        s_cmp_lt_u32      m0, s_save_alloc_size
-        s_cbranch_scc0    L_SAVE_VGPR_LOOP_END      // no more vgprs
 
-        s_set_gpr_idx_on  m0, 0x1   // This will change M0
-        s_add_u32         s_save_alloc_size, s_save_alloc_size, 0x1000  // because above inst change m0
-L_SAVE_VGPR_LOOP:
-        v_mov_b32         v0, v0   // v0 = v[0+m0]
-        v_mov_b32         v1, v1
-        v_mov_b32         v2, v2
-        v_mov_b32         v3, v3
-
-
-        buffer_store_dwordx4 v0, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1
-        s_add_u32         m0, m0, 4
-        s_add_u32         s_save_mem_offset, s_save_mem_offset, 256*4
-        s_cmp_lt_u32      m0, s_save_alloc_size
-    s_cbranch_scc1  L_SAVE_VGPR_LOOP                                                //VGPR save is complete?
-    s_set_gpr_idx_off
-L_SAVE_VGPR_LOOP_END:
-
-        s_and_b32 s_save_buf_rsrc1, s_save_buf_rsrc1, 0x0000FFFF   // reset const stride to 0
-        s_or_b32  s_save_buf_rsrc1, s_save_buf_rsrc1, S_SAVE_BUF_RSRC_WORD1_STRIDE  // reset const stride to 4 bytes
-else
     // VGPR store using dw burst
     s_mov_b32         m0, 0x4   //VGPR initial index value =0
     s_cmp_lt_u32      m0, s_save_alloc_size
@@ -700,52 +447,18 @@ else
     v_mov_b32       v2, v2              //v0 = v[0+m0]
     v_mov_b32       v3, v3              //v0 = v[0+m0]
 
-    if(USE_MTBUF_INSTEAD_OF_MUBUF)
-        tbuffer_store_format_x v0, v0, s_save_buf_rsrc0, s_save_mem_offset format:BUF_NUM_FORMAT_FLOAT format: BUF_DATA_FORMAT_32 slc:1 glc:1
-    else
         buffer_store_dword v0, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1
         buffer_store_dword v1, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1  offset:256
         buffer_store_dword v2, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1  offset:256*2
         buffer_store_dword v3, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1  offset:256*3
-    end
 
     s_add_u32       m0, m0, 4                                                       //next vgpr index
     s_add_u32       s_save_mem_offset, s_save_mem_offset, 256*4                     //every buffer_store_dword does 256 bytes
     s_cmp_lt_u32    m0, s_save_alloc_size                                           //scc = (m0 < s_save_alloc_size) ? 1 : 0
     s_cbranch_scc1  L_SAVE_VGPR_LOOP                                                //VGPR save is complete?
     s_set_gpr_idx_off
-end
 
 L_SAVE_VGPR_END:
-
-
-
-
-
-
-    /*     S_PGM_END_SAVED  */                              //FIXME  graphics ONLY
-    if ((EMU_RUN_HACK) && (!EMU_RUN_HACK_SAVE_NORMAL_EXIT))
-        s_and_b32 s_save_pc_hi, s_save_pc_hi, 0x0000ffff    //pc[47:32]
-        s_add_u32 s_save_pc_lo, s_save_pc_lo, 4             //pc[31:0]+4
-        s_addc_u32 s_save_pc_hi, s_save_pc_hi, 0x0          //carry bit over
-        s_rfe_b64 s_save_pc_lo                              //Return to the main shader program
-    else
-    end
-
-// Save Done timestamp
-if G8SR_DEBUG_TIMESTAMP
-        s_memrealtime   s_g8sr_ts_save_d
-        // SGPR SR memory offset : size(VGPR)
-        get_vgpr_size_bytes(s_save_mem_offset)
-        s_add_u32 s_save_mem_offset, s_save_mem_offset, G8SR_DEBUG_TS_SAVE_D_OFFSET
-        s_waitcnt lgkmcnt(0)         //FIXME, will cause xnack??
-        // Need reset rsrc2??
-        s_mov_b32 m0, s_save_mem_offset
-        s_mov_b32 s_save_buf_rsrc2,  0x1000000                                  //NUM_RECORDS in bytes
-        s_buffer_store_dwordx2 s_g8sr_ts_save_d, s_save_buf_rsrc0, m0       glc:1
-end
-
-
     s_branch    L_END_PGM
 
 
@@ -756,27 +469,6 @@ end
 
 L_RESTORE:
     /*      Setup Resource Contants    */
-    if ((EMU_RUN_HACK) && (!EMU_RUN_HACK_RESTORE_NORMAL))
-        //calculate wd_addr using absolute thread id
-        v_readlane_b32 s_restore_tmp, v9, 0
-        s_lshr_b32 s_restore_tmp, s_restore_tmp, 6
-        s_mul_i32 s_restore_tmp, s_restore_tmp, WAVE_SPACE
-        s_add_i32 s_restore_spi_init_lo, s_restore_tmp, WG_BASE_ADDR_LO
-        s_mov_b32 s_restore_spi_init_hi, WG_BASE_ADDR_HI
-        s_and_b32 s_restore_spi_init_hi, s_restore_spi_init_hi, CTX_RESTORE_CONTROL
-    else
-    end
-
-if G8SR_DEBUG_TIMESTAMP
-        s_memrealtime   s_g8sr_ts_restore_s
-        s_waitcnt lgkmcnt(0)         //FIXME, will cause xnack??
-        // tma_lo/hi are sgpr 110, 111, which will not used for 112 SGPR allocated case...
-        s_mov_b32 s_restore_pc_lo, s_g8sr_ts_restore_s[0]
-        s_mov_b32 s_restore_pc_hi, s_g8sr_ts_restore_s[1]   //backup ts to ttmp0/1, sicne exec will be finally restored..
-end
-
-
-
     s_mov_b32       s_restore_buf_rsrc0,    s_restore_spi_init_lo                                                           //base_addr_lo
     s_and_b32       s_restore_buf_rsrc1,    s_restore_spi_init_hi, 0x0000FFFF                                               //base_addr_hi
     s_or_b32        s_restore_buf_rsrc1,    s_restore_buf_rsrc1,  S_RESTORE_BUF_RSRC_WORD1_STRIDE
@@ -818,18 +510,12 @@ end
     s_add_u32  s_restore_mem_offset, s_restore_mem_offset, get_hwreg_size_bytes()            //FIXME, Check if offset overflow???
 
 
-    if (SWIZZLE_EN)
-        s_add_u32       s_restore_buf_rsrc2, s_restore_buf_rsrc2, 0x0                       //FIXME need to use swizzle to enable bounds checking?
-    else
         s_mov_b32       s_restore_buf_rsrc2,  0x1000000                                     //NUM_RECORDS in bytes
-    end
     s_mov_b32       m0, 0x0                                                                 //lds_offset initial value = 0
 
   L_RESTORE_LDS_LOOP:
-    if (SAVE_LDS)
         buffer_load_dword   v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset lds:1                    // first 64DW
         buffer_load_dword   v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset lds:1 offset:256         // second 64DW
-    end
     s_add_u32       m0, m0, 256*2                                               // 128 DW
     s_add_u32       s_restore_mem_offset, s_restore_mem_offset, 256*2           //mem offset increased by 128DW
     s_cmp_lt_u32    m0, s_restore_alloc_size                                    //scc=(m0 < s_restore_alloc_size) ? 1 : 0
@@ -848,40 +534,8 @@ end
     s_add_u32       s_restore_alloc_size, s_restore_alloc_size, 1
     s_lshl_b32      s_restore_alloc_size, s_restore_alloc_size, 2                           //Number of VGPRs = (vgpr_size + 1) * 4    (non-zero value)
     s_lshl_b32      s_restore_buf_rsrc2,  s_restore_alloc_size, 8                           //NUM_RECORDS in bytes (64 threads*4)
-    if (SWIZZLE_EN)
-        s_add_u32       s_restore_buf_rsrc2, s_restore_buf_rsrc2, 0x0                       //FIXME need to use swizzle to enable bounds checking?
-    else
         s_mov_b32       s_restore_buf_rsrc2,  0x1000000                                     //NUM_RECORDS in bytes
-    end
-
-if G8SR_VGPR_SR_IN_DWX4
-     get_vgpr_size_bytes(s_restore_mem_offset)
-     s_sub_u32         s_restore_mem_offset, s_restore_mem_offset, 256*4
-
-     // the const stride for DWx4 is 4*4 bytes
-     s_and_b32 s_restore_buf_rsrc1, s_restore_buf_rsrc1, 0x0000FFFF   // reset const stride to 0
-     s_or_b32  s_restore_buf_rsrc1, s_restore_buf_rsrc1, G8SR_RESTORE_BUF_RSRC_WORD1_STRIDE_DWx4  // const stride to 4*4 bytes
-
-     s_mov_b32         m0, s_restore_alloc_size
-     s_set_gpr_idx_on  m0, 0x8    // Note.. This will change m0
-
-L_RESTORE_VGPR_LOOP:
-     buffer_load_dwordx4 v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset slc:1 glc:1
-     s_waitcnt vmcnt(0)
-     s_sub_u32         m0, m0, 4
-     v_mov_b32         v0, v0   // v[0+m0] = v0
-     v_mov_b32         v1, v1
-     v_mov_b32         v2, v2
-     v_mov_b32         v3, v3
-     s_sub_u32         s_restore_mem_offset, s_restore_mem_offset, 256*4
-     s_cmp_eq_u32      m0, 0x8000
-     s_cbranch_scc0    L_RESTORE_VGPR_LOOP
-     s_set_gpr_idx_off
-
-     s_and_b32 s_restore_buf_rsrc1, s_restore_buf_rsrc1, 0x0000FFFF   // reset const stride to 0
-     s_or_b32  s_restore_buf_rsrc1, s_restore_buf_rsrc1, S_RESTORE_BUF_RSRC_WORD1_STRIDE  // const stride to 4*4 bytes
-
-else
+
     // VGPR load using dw burst
     s_mov_b32       s_restore_mem_offset_save, s_restore_mem_offset     // restore start with v1, v0 will be the last
     s_add_u32       s_restore_mem_offset, s_restore_mem_offset, 256*4
@@ -890,14 +544,10 @@ else
     s_add_u32       s_restore_alloc_size, s_restore_alloc_size, 0x8000                      //add 0x8000 since we compare m0 against it later
 
   L_RESTORE_VGPR_LOOP:
-    if(USE_MTBUF_INSTEAD_OF_MUBUF)
-        tbuffer_load_format_x v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset format:BUF_NUM_FORMAT_FLOAT format: BUF_DATA_FORMAT_32 slc:1 glc:1
-    else
         buffer_load_dword v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset slc:1 glc:1
         buffer_load_dword v1, v0, s_restore_buf_rsrc0, s_restore_mem_offset slc:1 glc:1 offset:256
         buffer_load_dword v2, v0, s_restore_buf_rsrc0, s_restore_mem_offset slc:1 glc:1 offset:256*2
         buffer_load_dword v3, v0, s_restore_buf_rsrc0, s_restore_mem_offset slc:1 glc:1 offset:256*3
-    end
     s_waitcnt       vmcnt(0)                                                                //ensure data ready
     v_mov_b32       v0, v0                                                                  //v[0+m0] = v0
     v_mov_b32       v1, v1
@@ -909,16 +559,10 @@ else
     s_cbranch_scc1  L_RESTORE_VGPR_LOOP                                                     //VGPR restore (except v0) is complete?
     s_set_gpr_idx_off
                                                                                             /* VGPR restore on v0 */
-    if(USE_MTBUF_INSTEAD_OF_MUBUF)
-        tbuffer_load_format_x v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save format:BUF_NUM_FORMAT_FLOAT format: BUF_DATA_FORMAT_32 slc:1 glc:1
-    else
         buffer_load_dword v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save    slc:1 glc:1
         buffer_load_dword v1, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save    slc:1 glc:1 offset:256
         buffer_load_dword v2, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save    slc:1 glc:1 offset:256*2
         buffer_load_dword v3, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save    slc:1 glc:1 offset:256*3
-    end
-
-end
 
     /*          restore SGPRs       */
     //////////////////////////////
@@ -934,16 +578,8 @@ end
     s_add_u32       s_restore_alloc_size, s_restore_alloc_size, 1
     s_lshl_b32      s_restore_alloc_size, s_restore_alloc_size, 4                           //Number of SGPRs = (sgpr_size + 1) * 16   (non-zero value)
 
-    if (SGPR_SAVE_USE_SQC)
         s_lshl_b32      s_restore_buf_rsrc2,    s_restore_alloc_size, 2                     //NUM_RECORDS in bytes
-    else
-        s_lshl_b32      s_restore_buf_rsrc2,    s_restore_alloc_size, 8                     //NUM_RECORDS in bytes (64 threads)
-    end
-    if (SWIZZLE_EN)
-        s_add_u32       s_restore_buf_rsrc2, s_restore_buf_rsrc2, 0x0                       //FIXME need to use swizzle to enable bounds checking?
-    else
         s_mov_b32       s_restore_buf_rsrc2,  0x1000000                                     //NUM_RECORDS in bytes
-    end
 
     /* If 112 SGPRs ar allocated, 4 sgprs are not used TBA(108,109),TMA(110,111),
        However, we are safe to restore these 4 SGPRs anyway, since TBA,TMA will later be restored by HWREG
@@ -972,12 +608,6 @@ end
     //////////////////////////////
   L_RESTORE_HWREG:
 
-
-if G8SR_DEBUG_TIMESTAMP
-      s_mov_b32 s_g8sr_ts_restore_s[0], s_restore_pc_lo
-      s_mov_b32 s_g8sr_ts_restore_s[1], s_restore_pc_hi
-end
-
     // HWREG SR memory offset : size(VGPR)+size(SGPR)
     get_vgpr_size_bytes(s_restore_mem_offset)
     get_sgpr_size_bytes(s_restore_tmp)
@@ -985,11 +615,7 @@ end
 
 
     s_mov_b32       s_restore_buf_rsrc2, 0x4                                                //NUM_RECORDS   in bytes
-    if (SWIZZLE_EN)
-        s_add_u32       s_restore_buf_rsrc2, s_restore_buf_rsrc2, 0x0                       //FIXME need to use swizzle to enable bounds checking?
-    else
         s_mov_b32       s_restore_buf_rsrc2,  0x1000000                                     //NUM_RECORDS in bytes
-    end
 
     read_hwreg_from_mem(s_restore_m0, s_restore_buf_rsrc0, s_restore_mem_offset)                    //M0
     read_hwreg_from_mem(s_restore_pc_lo, s_restore_buf_rsrc0, s_restore_mem_offset)             //PC
@@ -1006,16 +632,6 @@ end
 
     s_waitcnt       lgkmcnt(0)                                                                                      //from now on, it is safe to restore STATUS and IB_STS
 
-    //for normal save & restore, the saved PC points to the next inst to execute, no adjustment needs to be made, otherwise:
-    if ((EMU_RUN_HACK) && (!EMU_RUN_HACK_RESTORE_NORMAL))
-        s_add_u32 s_restore_pc_lo, s_restore_pc_lo, 8            //pc[31:0]+8     //two back-to-back s_trap are used (first for save and second for restore)
-        s_addc_u32  s_restore_pc_hi, s_restore_pc_hi, 0x0        //carry bit over
-    end
-    if ((EMU_RUN_HACK) && (EMU_RUN_HACK_RESTORE_NORMAL))
-        s_add_u32 s_restore_pc_lo, s_restore_pc_lo, 4            //pc[31:0]+4     // save is hack through s_trap but restore is normal
-        s_addc_u32  s_restore_pc_hi, s_restore_pc_hi, 0x0        //carry bit over
-    end
-
     s_mov_b32       m0,         s_restore_m0
     s_mov_b32       exec_lo,    s_restore_exec_lo
     s_mov_b32       exec_hi,    s_restore_exec_hi
@@ -1048,11 +664,6 @@ end
 
     s_barrier                                                   //barrier to ensure the readiness of LDS before access attempts from any other wave in the same TG //FIXME not performance-optimal at this time
 
-if G8SR_DEBUG_TIMESTAMP
-    s_memrealtime s_g8sr_ts_restore_d
-    s_waitcnt lgkmcnt(0)
-end
-
 //  s_rfe_b64 s_restore_pc_lo                                   //Return to the main shader program and resume execution
     s_rfe_restore_b64  s_restore_pc_lo, s_restore_m0            // s_restore_m0[0] is used to set STATUS.inst_atc
 
diff --git a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx9.asm b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx9.asm
index 6bae2e022c6e..75f29d13c90f 100644
--- a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx9.asm
+++ b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx9.asm
@@ -24,76 +24,9 @@
  * PROJECT=greenland ./sp3 cwsr_trap_handler_gfx9.asm -hex tmp.hex
  */
 
-/* HW (GFX9) source code for CWSR trap handler */
-/* Version 18 + multiple trap handler */
-
-// this performance-optimal version was originally from Seven Xu at SRDC
-
-// Revison #18	 --...
-/* Rev History
-** #1. Branch from gc dv.   //gfxip/gfx9/main/src/test/suites/block/cs/sr/cs_trap_handler.sp3#1,#50, #51, #52-53(Skip, Already Fixed by PV), #54-56(merged),#57-58(mergerd, skiped-already fixed by PV)
-** #4. SR Memory Layout:
-**			 1. VGPR-SGPR-HWREG-{LDS}
-**			 2. tba_hi.bits.26 - reconfigured as the first wave in tg bits, for defer Save LDS for a threadgroup.. performance concern..
-** #5. Update: 1. Accurate g8sr_ts_save_d timestamp
-** #6. Update: 1. Fix s_barrier usage; 2. VGPR s/r using swizzle buffer?(NoNeed, already matched the swizzle pattern, more investigation)
-** #7. Update: 1. don't barrier if noLDS
-** #8. Branch: 1. Branch to ver#0, which is very similar to gc dv version
-**	       2. Fix SQ issue by s_sleep 2
-** #9. Update: 1. Fix scc restore failed issue, restore wave_status at last
-**	       2. optimize s_buffer save by burst 16sgprs...
-** #10. Update 1. Optimize restore sgpr by busrt 16 sgprs.
-** #11. Update 1. Add 2 more timestamp for debug version
-** #12. Update 1. Add VGPR SR using DWx4, some case improve and some case drop performance
-** #13. Integ  1. Always use MUBUF for PV trap shader...
-** #14. Update 1. s_buffer_store soft clause...
-** #15. Update 1. PERF - sclar write with glc:0/mtype0 to allow L2 combine. perf improvement a lot.
-** #16. Update 1. PRRF - UNROLL LDS_DMA got 2500cycle save in IP tree
-** #17. Update 1. FUNC - LDS_DMA has issues while ATC, replace with ds_read/buffer_store for save part[TODO restore part]
-**	       2. PERF - Save LDS before save VGPR to cover LDS save long latency...
-** #18. Update 1. FUNC - Implicitly estore STATUS.VCCZ, which is not writable by s_setreg_b32
-**	       2. FUNC - Handle non-CWSR traps
-*/
-
-var G8SR_WDMEM_HWREG_OFFSET = 0
-var G8SR_WDMEM_SGPR_OFFSET  = 128  // in bytes
-
-// Keep definition same as the app shader, These 2 time stamps are part of the app shader... Should before any Save and after restore.
-
-var G8SR_DEBUG_TIMESTAMP = 0
-var G8SR_DEBUG_TS_SAVE_D_OFFSET = 40*4	// ts_save_d timestamp offset relative to SGPR_SR_memory_offset
-var s_g8sr_ts_save_s	= s[34:35]   // save start
-var s_g8sr_ts_sq_save_msg  = s[36:37]	// The save shader send SAVEWAVE msg to spi
-var s_g8sr_ts_spi_wrexec   = s[38:39]	// the SPI write the sr address to SQ
-var s_g8sr_ts_save_d	= s[40:41]   // save end
-var s_g8sr_ts_restore_s = s[42:43]   // restore start
-var s_g8sr_ts_restore_d = s[44:45]   // restore end
-
-var G8SR_VGPR_SR_IN_DWX4 = 0
-var G8SR_SAVE_BUF_RSRC_WORD1_STRIDE_DWx4 = 0x00100000	 // DWx4 stride is 4*4Bytes
-var G8SR_RESTORE_BUF_RSRC_WORD1_STRIDE_DWx4  = G8SR_SAVE_BUF_RSRC_WORD1_STRIDE_DWx4
-
-
-/*************************************************************************/
-/*		    control on how to run the shader			 */
-/*************************************************************************/
-//any hack that needs to be made to run this code in EMU (either because various EMU code are not ready or no compute save & restore in EMU run)
-var EMU_RUN_HACK		    =	0
-var EMU_RUN_HACK_RESTORE_NORMAL	    =	0
-var EMU_RUN_HACK_SAVE_NORMAL_EXIT   =	0
-var EMU_RUN_HACK_SAVE_SINGLE_WAVE   =	0
-var EMU_RUN_HACK_SAVE_FIRST_TIME    =	0		    //for interrupted restore in which the first save is through EMU_RUN_HACK
-var SAVE_LDS			    =	1
-var WG_BASE_ADDR_LO		    =	0x9000a000
-var WG_BASE_ADDR_HI		    =	0x0
-var WAVE_SPACE			    =	0x5000		    //memory size that each wave occupies in workgroup state mem
-var CTX_SAVE_CONTROL		    =	0x0
-var CTX_RESTORE_CONTROL		    =	CTX_SAVE_CONTROL
-var SIM_RUN_HACK		    =	0		    //any hack that needs to be made to run this code in SIM (either because various RTL code are not ready or no compute save & restore in RTL run)
-var SGPR_SAVE_USE_SQC		    =	1		    //use SQC D$ to do the write
-var USE_MTBUF_INSTEAD_OF_MUBUF	    =	0		    //because TC EMU currently asserts on 0 of // overload DFMT field to carry 4 more bits of stride for MUBUF opcodes
-var SWIZZLE_EN			    =	0		    //whether we use swizzled buffer addressing
 var ACK_SQC_STORE		    =	1		    //workaround for suspected SQC store bug causing incorrect stores under concurrency
+var SAVE_AFTER_XNACK_ERROR	    =	1		    //workaround for TCP store failure after XNACK error when ALLOW_REPLAY=0, for debugger
+var SINGLE_STEP_MISSED_WORKAROUND   =	1		    //workaround for lost MODE.DEBUG_EN exception when SAVECTX raised
 
 /**************************************************************************/
 /*			variables					  */
@@ -107,6 +40,7 @@ var SQ_WAVE_STATUS_PRE_SPI_PRIO_SHIFT   = 0
 var SQ_WAVE_STATUS_PRE_SPI_PRIO_SIZE    = 1
 var SQ_WAVE_STATUS_POST_SPI_PRIO_SHIFT  = 3
 var SQ_WAVE_STATUS_POST_SPI_PRIO_SIZE   = 29
+var SQ_WAVE_STATUS_ALLOW_REPLAY_MASK    = 0x400000
 
 var SQ_WAVE_LDS_ALLOC_LDS_SIZE_SHIFT	= 12
 var SQ_WAVE_LDS_ALLOC_LDS_SIZE_SIZE	= 9
@@ -127,12 +61,15 @@ var SQ_WAVE_TRAPSTS_POST_SAVECTX_MASK	=   0xFFFFF800
 var SQ_WAVE_TRAPSTS_POST_SAVECTX_SHIFT	=   11
 var SQ_WAVE_TRAPSTS_POST_SAVECTX_SIZE	=   21
 var SQ_WAVE_TRAPSTS_ILLEGAL_INST_MASK	=   0x800
+var SQ_WAVE_TRAPSTS_XNACK_ERROR_MASK	=   0x10000000
 
 var SQ_WAVE_IB_STS_RCNT_SHIFT		=   16			//FIXME
 var SQ_WAVE_IB_STS_FIRST_REPLAY_SHIFT	=   15			//FIXME
 var SQ_WAVE_IB_STS_RCNT_FIRST_REPLAY_MASK	= 0x1F8000
 var SQ_WAVE_IB_STS_RCNT_FIRST_REPLAY_MASK_NEG	= 0x00007FFF	//FIXME
 
+var SQ_WAVE_MODE_DEBUG_EN_MASK		=   0x800
+
 var SQ_BUF_RSRC_WORD1_ATC_SHIFT	    =	24
 var SQ_BUF_RSRC_WORD3_MTYPE_SHIFT   =	27
 
@@ -197,13 +134,15 @@ var s_restore_spi_init_lo		    =	exec_lo
 var s_restore_spi_init_hi		    =	exec_hi
 
 var s_restore_mem_offset	=   ttmp12
+var s_restore_accvgpr_offset	=   ttmp13
 var s_restore_alloc_size	=   ttmp3
 var s_restore_tmp		=   ttmp2
 var s_restore_mem_offset_save	=   s_restore_tmp	//no conflict
+var s_restore_accvgpr_offset_save = ttmp7
 
 var s_restore_m0	    =	s_restore_alloc_size	//no conflict
 
-var s_restore_mode	    =	ttmp7
+var s_restore_mode	    =	s_restore_accvgpr_offset_save
 
 var s_restore_pc_lo	    =	ttmp0
 var s_restore_pc_hi	    =	ttmp1
@@ -226,20 +165,11 @@ var s_restore_ttmps_hi	    =	s_restore_alloc_size	//no conflict
 /* Shader Main*/
 
 shader main
-  asic(GFX9)
+  asic(DEFAULT)
   type(CS)
 
 
-    if ((EMU_RUN_HACK) && (!EMU_RUN_HACK_RESTORE_NORMAL))		    //hack to use trap_id for determining save/restore
-	//FIXME VCCZ un-init assertion s_getreg_b32	s_save_status, hwreg(HW_REG_STATUS)	    //save STATUS since we will change SCC
-	s_and_b32 s_save_tmp, s_save_pc_hi, 0xffff0000		    //change SCC
-	s_cmp_eq_u32 s_save_tmp, 0x007e0000			    //Save: trap_id = 0x7e. Restore: trap_id = 0x7f.
-	s_cbranch_scc0 L_JUMP_TO_RESTORE			    //do not need to recover STATUS here  since we are going to RESTORE
-	//FIXME	 s_setreg_b32	hwreg(HW_REG_STATUS),	s_save_status	    //need to recover STATUS since we are going to SAVE
-	s_branch L_SKIP_RESTORE					    //NOT restore, SAVE actually
-    else
 	s_branch L_SKIP_RESTORE					    //NOT restore. might be a regular trap or save
-    end
 
 L_JUMP_TO_RESTORE:
     s_branch L_RESTORE						    //restore
@@ -248,12 +178,29 @@ L_SKIP_RESTORE:
 
     s_getreg_b32    s_save_status, hwreg(HW_REG_STATUS)				    //save STATUS since we will change SCC
     s_andn2_b32	    s_save_status, s_save_status, SQ_WAVE_STATUS_SPI_PRIO_MASK	    //check whether this is for save
+
+if SINGLE_STEP_MISSED_WORKAROUND
+    // No single step exceptions if MODE.DEBUG_EN=0.
+    s_getreg_b32    ttmp2, hwreg(HW_REG_MODE)
+    s_and_b32       ttmp2, ttmp2, SQ_WAVE_MODE_DEBUG_EN_MASK
+    s_cbranch_scc0  L_NO_SINGLE_STEP_WORKAROUND
+
+    // Second-level trap already handled exception if STATUS.HALT=1.
+    s_and_b32       ttmp2, s_save_status, SQ_WAVE_STATUS_HALT_MASK
+
+    // Prioritize single step exception over context save.
+    // Second-level trap will halt wave and RFE, re-entering for SAVECTX.
+    s_cbranch_scc0  L_FETCH_2ND_TRAP
+
+L_NO_SINGLE_STEP_WORKAROUND:
+end
+
     s_getreg_b32    s_save_trapsts, hwreg(HW_REG_TRAPSTS)
     s_and_b32       ttmp2, s_save_trapsts, SQ_WAVE_TRAPSTS_SAVECTX_MASK    //check whether this is for save
     s_cbranch_scc1  L_SAVE					//this is the operation for save
 
     // *********    Handle non-CWSR traps	*******************
-if (!EMU_RUN_HACK)
+
     // Illegal instruction is a non-maskable exception which blocks context save.
     // Halt the wavefront and return from the trap.
     s_and_b32       ttmp2, s_save_trapsts, SQ_WAVE_TRAPSTS_ILLEGAL_INST_MASK
@@ -330,7 +277,7 @@ L_EXCP_CASE:
     set_status_without_spi_prio(s_save_status, ttmp2)
 
     s_rfe_b64       [ttmp0, ttmp1]
-end
+
     // *********	End handling of non-CWSR traps	 *******************
 
 /**************************************************************************/
@@ -338,12 +285,6 @@ end
 /**************************************************************************/
 
 L_SAVE:
-
-if G8SR_DEBUG_TIMESTAMP
-	s_memrealtime	s_g8sr_ts_save_s
-	s_waitcnt lgkmcnt(0)	     //FIXME, will cause xnack??
-end
-
     s_and_b32	    s_save_pc_hi, s_save_pc_hi, 0x0000ffff    //pc[47:32]
 
     s_mov_b32	    s_save_tmp, 0							    //clear saveCtx bit
@@ -365,16 +306,7 @@ end
     s_mov_b32	    s_save_exec_hi, exec_hi
     s_mov_b64	    exec,   0x0								    //clear EXEC to get ready to receive
 
-if G8SR_DEBUG_TIMESTAMP
-	s_memrealtime  s_g8sr_ts_sq_save_msg
-	s_waitcnt lgkmcnt(0)
-end
-
-    if (EMU_RUN_HACK)
-
-    else
 	s_sendmsg   sendmsg(MSG_SAVEWAVE)  //send SPI a message and wait for SPI's write to EXEC
-    end
 
     // Set SPI_PRIO=2 to avoid starving instruction fetch in the waves we're waiting for.
     s_or_b32 s_save_tmp, s_save_status, (2 << SQ_WAVE_STATUS_SPI_PRIO_SHIFT)
@@ -383,33 +315,7 @@ end
   L_SLEEP:
     s_sleep 0x2		       // sleep 1 (64clk) is not enough for 8 waves per SIMD, which will cause SQ hang, since the 7,8th wave could not get arbit to exec inst, while other waves are stuck into the sleep-loop and waiting for wrexec!=0
 
-    if (EMU_RUN_HACK)
-
-    else
 	s_cbranch_execz L_SLEEP
-    end
-
-if G8SR_DEBUG_TIMESTAMP
-	s_memrealtime  s_g8sr_ts_spi_wrexec
-	s_waitcnt lgkmcnt(0)
-end
-
-    if ((EMU_RUN_HACK) && (!EMU_RUN_HACK_SAVE_SINGLE_WAVE))
-	//calculate wd_addr using absolute thread id
-	v_readlane_b32 s_save_tmp, v9, 0
-	s_lshr_b32 s_save_tmp, s_save_tmp, 6
-	s_mul_i32 s_save_tmp, s_save_tmp, WAVE_SPACE
-	s_add_i32 s_save_spi_init_lo, s_save_tmp, WG_BASE_ADDR_LO
-	s_mov_b32 s_save_spi_init_hi, WG_BASE_ADDR_HI
-	s_and_b32 s_save_spi_init_hi, s_save_spi_init_hi, CTX_SAVE_CONTROL
-    else
-    end
-    if ((EMU_RUN_HACK) && (EMU_RUN_HACK_SAVE_SINGLE_WAVE))
-	s_add_i32 s_save_spi_init_lo, s_save_tmp, WG_BASE_ADDR_LO
-	s_mov_b32 s_save_spi_init_hi, WG_BASE_ADDR_HI
-	s_and_b32 s_save_spi_init_hi, s_save_spi_init_hi, CTX_SAVE_CONTROL
-    else
-    end
 
     // Save trap temporaries 4-11, 13 initialized by SPI debug dispatch logic
     // ttmp SR memory offset : size(VGPR)+size(SGPR)+0x40
@@ -459,20 +365,10 @@ end
 
 
     s_mov_b32	    s_save_buf_rsrc2, 0x4				//NUM_RECORDS	in bytes
-    if (SWIZZLE_EN)
-	s_add_u32	s_save_buf_rsrc2, s_save_buf_rsrc2, 0x0			    //FIXME need to use swizzle to enable bounds checking?
-    else
 	s_mov_b32	s_save_buf_rsrc2,  0x1000000				    //NUM_RECORDS in bytes
-    end
 
 
     write_hwreg_to_mem(s_save_m0, s_save_buf_rsrc0, s_save_mem_offset)			//M0
-
-    if ((EMU_RUN_HACK) && (EMU_RUN_HACK_SAVE_FIRST_TIME))
-	s_add_u32 s_save_pc_lo, s_save_pc_lo, 4		    //pc[31:0]+4
-	s_addc_u32 s_save_pc_hi, s_save_pc_hi, 0x0	    //carry bit over
-    end
-
     write_hwreg_to_mem(s_save_pc_lo, s_save_buf_rsrc0, s_save_mem_offset)		    //PC
     write_hwreg_to_mem(s_save_pc_hi, s_save_buf_rsrc0, s_save_mem_offset)
     write_hwreg_to_mem(s_save_exec_lo, s_save_buf_rsrc0, s_save_mem_offset)		//EXEC
@@ -510,17 +406,9 @@ end
     s_add_u32	    s_save_alloc_size, s_save_alloc_size, 1
     s_lshl_b32	    s_save_alloc_size, s_save_alloc_size, 4			    //Number of SGPRs = (sgpr_size + 1) * 16   (non-zero value)
 
-    if (SGPR_SAVE_USE_SQC)
 	s_lshl_b32	s_save_buf_rsrc2,   s_save_alloc_size, 2		    //NUM_RECORDS in bytes
-    else
-	s_lshl_b32	s_save_buf_rsrc2,   s_save_alloc_size, 8		    //NUM_RECORDS in bytes (64 threads)
-    end
 
-    if (SWIZZLE_EN)
-	s_add_u32	s_save_buf_rsrc2, s_save_buf_rsrc2, 0x0			    //FIXME need to use swizzle to enable bounds checking?
-    else
 	s_mov_b32	s_save_buf_rsrc2,  0x1000000				    //NUM_RECORDS in bytes
-    end
 
 
     // backup s_save_buf_rsrc0,1 to s_save_pc_lo/hi, since write_16sgpr_to_mem function will change the rsrc0
@@ -563,30 +451,25 @@ end
     s_mov_b32	    xnack_mask_lo, 0x0
     s_mov_b32	    xnack_mask_hi, 0x0
 
-    if (SWIZZLE_EN)
-	s_add_u32	s_save_buf_rsrc2, s_save_buf_rsrc2, 0x0			    //FIXME need to use swizzle to enable bounds checking?
-    else
 	s_mov_b32	s_save_buf_rsrc2,  0x1000000				    //NUM_RECORDS in bytes
-    end
 
 
     // VGPR Allocated in 4-GPR granularity
 
-if G8SR_VGPR_SR_IN_DWX4
-	// the const stride for DWx4 is 4*4 bytes
-	s_and_b32 s_save_buf_rsrc1, s_save_buf_rsrc1, 0x0000FFFF   // reset const stride to 0
-	s_or_b32  s_save_buf_rsrc1, s_save_buf_rsrc1, G8SR_SAVE_BUF_RSRC_WORD1_STRIDE_DWx4  // const stride to 4*4 bytes
+if SAVE_AFTER_XNACK_ERROR
+	check_if_tcp_store_ok()
+	s_cbranch_scc1 L_SAVE_FIRST_VGPRS_WITH_TCP
+
+	write_vgprs_to_mem_with_sqc(v0, 4, s_save_buf_rsrc0, s_save_mem_offset)
+	s_branch L_SAVE_LDS
 
-	buffer_store_dwordx4 v0, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1
+L_SAVE_FIRST_VGPRS_WITH_TCP:
+end
 
-	s_and_b32 s_save_buf_rsrc1, s_save_buf_rsrc1, 0x0000FFFF   // reset const stride to 0
-	s_or_b32  s_save_buf_rsrc1, s_save_buf_rsrc1, S_SAVE_BUF_RSRC_WORD1_STRIDE  // reset const stride to 4 bytes
-else
 	buffer_store_dword v0, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1
 	buffer_store_dword v1, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1  offset:256
 	buffer_store_dword v2, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1  offset:256*2
 	buffer_store_dword v3, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1  offset:256*3
-end
 
 
 
@@ -621,66 +504,34 @@ end
     s_add_u32 s_save_mem_offset, s_save_mem_offset, get_hwreg_size_bytes()
 
 
-    if (SWIZZLE_EN)
-	s_add_u32	s_save_buf_rsrc2, s_save_buf_rsrc2, 0x0	      //FIXME need to use swizzle to enable bounds checking?
-    else
 	s_mov_b32	s_save_buf_rsrc2,  0x1000000		      //NUM_RECORDS in bytes
-    end
 
     s_mov_b32	    m0, 0x0						  //lds_offset initial value = 0
 
 
-var LDS_DMA_ENABLE = 0
-var UNROLL = 0
-if UNROLL==0 && LDS_DMA_ENABLE==1
-	s_mov_b32  s3, 256*2
-	s_nop 0
-	s_nop 0
-	s_nop 0
-  L_SAVE_LDS_LOOP:
-	//TODO: looks the 2 buffer_store/load clause for s/r will hurt performance.???
-    if (SAVE_LDS)     //SPI always alloc LDS space in 128DW granularity
-	    buffer_store_lds_dword s_save_buf_rsrc0, s_save_mem_offset lds:1		// first 64DW
-	    buffer_store_lds_dword s_save_buf_rsrc0, s_save_mem_offset lds:1 offset:256 // second 64DW
-    end
-
-    s_add_u32	    m0, m0, s3						//every buffer_store_lds does 256 bytes
-    s_add_u32	    s_save_mem_offset, s_save_mem_offset, s3				//mem offset increased by 256 bytes
-    s_cmp_lt_u32    m0, s_save_alloc_size						//scc=(m0 < s_save_alloc_size) ? 1 : 0
-    s_cbranch_scc1  L_SAVE_LDS_LOOP							//LDS save is complete?
-
-elsif LDS_DMA_ENABLE==1 && UNROLL==1 // UNROOL	, has ichace miss
-      // store from higest LDS address to lowest
-      s_mov_b32	 s3, 256*2
-      s_sub_u32	 m0, s_save_alloc_size, s3
-      s_add_u32 s_save_mem_offset, s_save_mem_offset, m0
-      s_lshr_b32 s_save_alloc_size, s_save_alloc_size, 9   // how many 128 trunks...
-      s_sub_u32 s_save_alloc_size, 128, s_save_alloc_size   // store from higheset addr to lowest
-      s_mul_i32 s_save_alloc_size, s_save_alloc_size, 6*4   // PC offset increment,  each LDS save block cost 6*4 Bytes instruction
-      s_add_u32 s_save_alloc_size, s_save_alloc_size, 3*4   //2is the below 2 inst...//s_addc and s_setpc
-      s_nop 0
-      s_nop 0
-      s_nop 0	//pad 3 dw to let LDS_DMA align with 64Bytes
-      s_getpc_b64 s[0:1]			      // reuse s[0:1], since s[0:1] already saved
-      s_add_u32	  s0, s0,s_save_alloc_size
-      s_addc_u32  s1, s1, 0
-      s_setpc_b64 s[0:1]
-
-
-       for var i =0; i< 128; i++
-	    // be careful to make here a 64Byte aligned address, which could improve performance...
-	    buffer_store_lds_dword s_save_buf_rsrc0, s_save_mem_offset lds:1 offset:0		// first 64DW
-	    buffer_store_lds_dword s_save_buf_rsrc0, s_save_mem_offset lds:1 offset:256		  // second 64DW
-
-	if i!=127
-	s_sub_u32  m0, m0, s3	   // use a sgpr to shrink 2DW-inst to 1DW inst to improve performance , i.e.  pack more LDS_DMA inst to one Cacheline
-	    s_sub_u32  s_save_mem_offset, s_save_mem_offset,  s3
-	    end
-       end
-
-else   // BUFFER_STORE
       v_mbcnt_lo_u32_b32 v2, 0xffffffff, 0x0
       v_mbcnt_hi_u32_b32 v3, 0xffffffff, v2	// tid
+
+if SAVE_AFTER_XNACK_ERROR
+	check_if_tcp_store_ok()
+	s_cbranch_scc1 L_SAVE_LDS_WITH_TCP
+
+	v_lshlrev_b32 v2, 2, v3
+L_SAVE_LDS_LOOP_SQC:
+	ds_read2_b32 v[0:1], v2 offset0:0 offset1:0x40
+	s_waitcnt lgkmcnt(0)
+
+	write_vgprs_to_mem_with_sqc(v0, 2, s_save_buf_rsrc0, s_save_mem_offset)
+
+	v_add_u32 v2, 0x200, v2
+	v_cmp_lt_u32 vcc[0:1], v2, s_save_alloc_size
+	s_cbranch_vccnz L_SAVE_LDS_LOOP_SQC
+
+	s_branch L_SAVE_LDS_DONE
+
+L_SAVE_LDS_WITH_TCP:
+end
+
       v_mul_i32_i24 v2, v3, 8	// tid*8
       v_mov_b32 v3, 256*2
       s_mov_b32 m0, 0x10000
@@ -701,8 +552,6 @@ L_SAVE_LDS_LOOP_VECTOR:
       // restore rsrc3
       s_mov_b32 s_save_buf_rsrc3, s0
 
-end
-
 L_SAVE_LDS_DONE:
 
 
@@ -720,44 +569,9 @@ L_SAVE_LDS_DONE:
     s_add_u32	    s_save_alloc_size, s_save_alloc_size, 1
     s_lshl_b32	    s_save_alloc_size, s_save_alloc_size, 2			    //Number of VGPRs = (vgpr_size + 1) * 4    (non-zero value)	  //FIXME for GFX, zero is possible
     s_lshl_b32	    s_save_buf_rsrc2,  s_save_alloc_size, 8			    //NUM_RECORDS in bytes (64 threads*4)
-    if (SWIZZLE_EN)
-	s_add_u32	s_save_buf_rsrc2, s_save_buf_rsrc2, 0x0			    //FIXME need to use swizzle to enable bounds checking?
-    else
 	s_mov_b32	s_save_buf_rsrc2,  0x1000000				    //NUM_RECORDS in bytes
-    end
-
-
-    // VGPR Allocated in 4-GPR granularity
-
-if G8SR_VGPR_SR_IN_DWX4
-	// the const stride for DWx4 is 4*4 bytes
-	s_and_b32 s_save_buf_rsrc1, s_save_buf_rsrc1, 0x0000FFFF   // reset const stride to 0
-	s_or_b32  s_save_buf_rsrc1, s_save_buf_rsrc1, G8SR_SAVE_BUF_RSRC_WORD1_STRIDE_DWx4  // const stride to 4*4 bytes
-
-	s_mov_b32	  m0, 4	    // skip first 4 VGPRs
-	s_cmp_lt_u32	  m0, s_save_alloc_size
-	s_cbranch_scc0	  L_SAVE_VGPR_LOOP_END	    // no more vgprs
-
-	s_set_gpr_idx_on  m0, 0x1   // This will change M0
-	s_add_u32	  s_save_alloc_size, s_save_alloc_size, 0x1000	// because above inst change m0
-L_SAVE_VGPR_LOOP:
-	v_mov_b32	  v0, v0   // v0 = v[0+m0]
-	v_mov_b32	  v1, v1
-	v_mov_b32	  v2, v2
-	v_mov_b32	  v3, v3
-
 
-	buffer_store_dwordx4 v0, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1
-	s_add_u32	  m0, m0, 4
-	s_add_u32	  s_save_mem_offset, s_save_mem_offset, 256*4
-	s_cmp_lt_u32	  m0, s_save_alloc_size
-    s_cbranch_scc1  L_SAVE_VGPR_LOOP						    //VGPR save is complete?
-    s_set_gpr_idx_off
-L_SAVE_VGPR_LOOP_END:
 
-	s_and_b32 s_save_buf_rsrc1, s_save_buf_rsrc1, 0x0000FFFF   // reset const stride to 0
-	s_or_b32  s_save_buf_rsrc1, s_save_buf_rsrc1, S_SAVE_BUF_RSRC_WORD1_STRIDE  // reset const stride to 4 bytes
-else
     // VGPR store using dw burst
     s_mov_b32	      m0, 0x4	//VGPR initial index value =0
     s_cmp_lt_u32      m0, s_save_alloc_size
@@ -767,57 +581,82 @@ else
     s_set_gpr_idx_on	m0, 0x1 //M0[7:0] = M0[7:0] and M0[15:12] = 0x1
     s_add_u32	    s_save_alloc_size, s_save_alloc_size, 0x1000		    //add 0x1000 since we compare m0 against it later
 
+if SAVE_AFTER_XNACK_ERROR
+	check_if_tcp_store_ok()
+	s_cbranch_scc1 L_SAVE_VGPR_LOOP
+
+L_SAVE_VGPR_LOOP_SQC:
+	write_vgprs_to_mem_with_sqc(v0, 4, s_save_buf_rsrc0, s_save_mem_offset)
+
+	s_add_u32 m0, m0, 4
+	s_cmp_lt_u32 m0, s_save_alloc_size
+	s_cbranch_scc1 L_SAVE_VGPR_LOOP_SQC
+
+	s_set_gpr_idx_off
+	s_branch L_SAVE_VGPR_END
+end
+
   L_SAVE_VGPR_LOOP:
     v_mov_b32	    v0, v0		//v0 = v[0+m0]
     v_mov_b32	    v1, v1		//v0 = v[0+m0]
     v_mov_b32	    v2, v2		//v0 = v[0+m0]
     v_mov_b32	    v3, v3		//v0 = v[0+m0]
 
-    if(USE_MTBUF_INSTEAD_OF_MUBUF)
-	tbuffer_store_format_x v0, v0, s_save_buf_rsrc0, s_save_mem_offset format:BUF_NUM_FORMAT_FLOAT format: BUF_DATA_FORMAT_32 slc:1 glc:1
-    else
 	buffer_store_dword v0, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1
 	buffer_store_dword v1, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1  offset:256
 	buffer_store_dword v2, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1  offset:256*2
 	buffer_store_dword v3, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1  offset:256*3
-    end
 
     s_add_u32	    m0, m0, 4							    //next vgpr index
     s_add_u32	    s_save_mem_offset, s_save_mem_offset, 256*4			    //every buffer_store_dword does 256 bytes
     s_cmp_lt_u32    m0, s_save_alloc_size					    //scc = (m0 < s_save_alloc_size) ? 1 : 0
     s_cbranch_scc1  L_SAVE_VGPR_LOOP						    //VGPR save is complete?
     s_set_gpr_idx_off
-end
 
 L_SAVE_VGPR_END:
 
+if ASIC_TARGET_ARCTURUS
+    // Save ACC VGPRs
+    s_mov_b32 m0, 0x0 //VGPR initial index value =0
+    s_set_gpr_idx_on m0, 0x1 //M0[7:0] = M0[7:0] and M0[15:12] = 0x1
+
+if SAVE_AFTER_XNACK_ERROR
+    check_if_tcp_store_ok()
+    s_cbranch_scc1 L_SAVE_ACCVGPR_LOOP
 
+L_SAVE_ACCVGPR_LOOP_SQC:
+    for var vgpr = 0; vgpr < 4; ++ vgpr
+        v_accvgpr_read v[vgpr], acc[vgpr]  // v[N] = acc[N+m0]
+    end
 
+    write_vgprs_to_mem_with_sqc(v0, 4, s_save_buf_rsrc0, s_save_mem_offset)
 
+    s_add_u32 m0, m0, 4
+    s_cmp_lt_u32 m0, s_save_alloc_size
+    s_cbranch_scc1 L_SAVE_ACCVGPR_LOOP_SQC
 
+    s_set_gpr_idx_off
+    s_branch L_SAVE_ACCVGPR_END
+end
 
-    /*	   S_PGM_END_SAVED  */				    //FIXME  graphics ONLY
-    if ((EMU_RUN_HACK) && (!EMU_RUN_HACK_SAVE_NORMAL_EXIT))
-	s_and_b32 s_save_pc_hi, s_save_pc_hi, 0x0000ffff    //pc[47:32]
-	s_add_u32 s_save_pc_lo, s_save_pc_lo, 4		    //pc[31:0]+4
-	s_addc_u32 s_save_pc_hi, s_save_pc_hi, 0x0	    //carry bit over
-	s_rfe_b64 s_save_pc_lo				    //Return to the main shader program
-    else
+L_SAVE_ACCVGPR_LOOP:
+    for var vgpr = 0; vgpr < 4; ++ vgpr
+        v_accvgpr_read v[vgpr], acc[vgpr]  // v[N] = acc[N+m0]
     end
 
-// Save Done timestamp
-if G8SR_DEBUG_TIMESTAMP
-	s_memrealtime	s_g8sr_ts_save_d
-	// SGPR SR memory offset : size(VGPR)
-	get_vgpr_size_bytes(s_save_mem_offset)
-	s_add_u32 s_save_mem_offset, s_save_mem_offset, G8SR_DEBUG_TS_SAVE_D_OFFSET
-	s_waitcnt lgkmcnt(0)	     //FIXME, will cause xnack??
-	// Need reset rsrc2??
-	s_mov_b32 m0, s_save_mem_offset
-	s_mov_b32 s_save_buf_rsrc2,  0x1000000					//NUM_RECORDS in bytes
-	s_buffer_store_dwordx2 s_g8sr_ts_save_d, s_save_buf_rsrc0, m0	    glc:1
-end
+    buffer_store_dword v0, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1
+    buffer_store_dword v1, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1 offset:256
+    buffer_store_dword v2, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1 offset:256*2
+    buffer_store_dword v3, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1 offset:256*3
 
+    s_add_u32 m0, m0, 4
+    s_add_u32 s_save_mem_offset, s_save_mem_offset, 256*4
+    s_cmp_lt_u32 m0, s_save_alloc_size
+    s_cbranch_scc1 L_SAVE_ACCVGPR_LOOP
+    s_set_gpr_idx_off
+
+L_SAVE_ACCVGPR_END:
+end
 
     s_branch	L_END_PGM
 
@@ -829,27 +668,6 @@ end
 
 L_RESTORE:
     /*	    Setup Resource Contants    */
-    if ((EMU_RUN_HACK) && (!EMU_RUN_HACK_RESTORE_NORMAL))
-	//calculate wd_addr using absolute thread id
-	v_readlane_b32 s_restore_tmp, v9, 0
-	s_lshr_b32 s_restore_tmp, s_restore_tmp, 6
-	s_mul_i32 s_restore_tmp, s_restore_tmp, WAVE_SPACE
-	s_add_i32 s_restore_spi_init_lo, s_restore_tmp, WG_BASE_ADDR_LO
-	s_mov_b32 s_restore_spi_init_hi, WG_BASE_ADDR_HI
-	s_and_b32 s_restore_spi_init_hi, s_restore_spi_init_hi, CTX_RESTORE_CONTROL
-    else
-    end
-
-if G8SR_DEBUG_TIMESTAMP
-	s_memrealtime	s_g8sr_ts_restore_s
-	s_waitcnt lgkmcnt(0)	     //FIXME, will cause xnack??
-	// tma_lo/hi are sgpr 110, 111, which will not used for 112 SGPR allocated case...
-	s_mov_b32 s_restore_pc_lo, s_g8sr_ts_restore_s[0]
-	s_mov_b32 s_restore_pc_hi, s_g8sr_ts_restore_s[1]   //backup ts to ttmp0/1, sicne exec will be finally restored..
-end
-
-
-
     s_mov_b32	    s_restore_buf_rsrc0,    s_restore_spi_init_lo							    //base_addr_lo
     s_and_b32	    s_restore_buf_rsrc1,    s_restore_spi_init_hi, 0x0000FFFF						    //base_addr_hi
     s_or_b32	    s_restore_buf_rsrc1,    s_restore_buf_rsrc1,  S_RESTORE_BUF_RSRC_WORD1_STRIDE
@@ -891,18 +709,12 @@ end
     s_add_u32  s_restore_mem_offset, s_restore_mem_offset, get_hwreg_size_bytes()	     //FIXME, Check if offset overflow???
 
 
-    if (SWIZZLE_EN)
-	s_add_u32	s_restore_buf_rsrc2, s_restore_buf_rsrc2, 0x0			    //FIXME need to use swizzle to enable bounds checking?
-    else
 	s_mov_b32	s_restore_buf_rsrc2,  0x1000000					    //NUM_RECORDS in bytes
-    end
     s_mov_b32	    m0, 0x0								    //lds_offset initial value = 0
 
   L_RESTORE_LDS_LOOP:
-    if (SAVE_LDS)
 	buffer_load_dword   v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset lds:1		       // first 64DW
 	buffer_load_dword   v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset lds:1 offset:256	       // second 64DW
-    end
     s_add_u32	    m0, m0, 256*2						// 128 DW
     s_add_u32	    s_restore_mem_offset, s_restore_mem_offset, 256*2		//mem offset increased by 128DW
     s_cmp_lt_u32    m0, s_restore_alloc_size					//scc=(m0 < s_restore_alloc_size) ? 1 : 0
@@ -921,56 +733,43 @@ end
     s_add_u32	    s_restore_alloc_size, s_restore_alloc_size, 1
     s_lshl_b32	    s_restore_alloc_size, s_restore_alloc_size, 2			    //Number of VGPRs = (vgpr_size + 1) * 4    (non-zero value)
     s_lshl_b32	    s_restore_buf_rsrc2,  s_restore_alloc_size, 8			    //NUM_RECORDS in bytes (64 threads*4)
-    if (SWIZZLE_EN)
-	s_add_u32	s_restore_buf_rsrc2, s_restore_buf_rsrc2, 0x0			    //FIXME need to use swizzle to enable bounds checking?
-    else
+
+if ASIC_TARGET_ARCTURUS
+    s_mov_b32	    s_restore_accvgpr_offset, s_restore_buf_rsrc2                           //ACC VGPRs at end of VGPRs
+end
+
 	s_mov_b32	s_restore_buf_rsrc2,  0x1000000					    //NUM_RECORDS in bytes
-    end
 
-if G8SR_VGPR_SR_IN_DWX4
-     get_vgpr_size_bytes(s_restore_mem_offset)
-     s_sub_u32	       s_restore_mem_offset, s_restore_mem_offset, 256*4
-
-     // the const stride for DWx4 is 4*4 bytes
-     s_and_b32 s_restore_buf_rsrc1, s_restore_buf_rsrc1, 0x0000FFFF   // reset const stride to 0
-     s_or_b32  s_restore_buf_rsrc1, s_restore_buf_rsrc1, G8SR_RESTORE_BUF_RSRC_WORD1_STRIDE_DWx4  // const stride to 4*4 bytes
-
-     s_mov_b32	       m0, s_restore_alloc_size
-     s_set_gpr_idx_on  m0, 0x8	  // Note.. This will change m0
-
-L_RESTORE_VGPR_LOOP:
-     buffer_load_dwordx4 v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset slc:1 glc:1
-     s_waitcnt vmcnt(0)
-     s_sub_u32	       m0, m0, 4
-     v_mov_b32	       v0, v0	// v[0+m0] = v0
-     v_mov_b32	       v1, v1
-     v_mov_b32	       v2, v2
-     v_mov_b32	       v3, v3
-     s_sub_u32	       s_restore_mem_offset, s_restore_mem_offset, 256*4
-     s_cmp_eq_u32      m0, 0x8000
-     s_cbranch_scc0    L_RESTORE_VGPR_LOOP
-     s_set_gpr_idx_off
-
-     s_and_b32 s_restore_buf_rsrc1, s_restore_buf_rsrc1, 0x0000FFFF   // reset const stride to 0
-     s_or_b32  s_restore_buf_rsrc1, s_restore_buf_rsrc1, S_RESTORE_BUF_RSRC_WORD1_STRIDE  // const stride to 4*4 bytes
-
-else
     // VGPR load using dw burst
     s_mov_b32	    s_restore_mem_offset_save, s_restore_mem_offset	// restore start with v1, v0 will be the last
     s_add_u32	    s_restore_mem_offset, s_restore_mem_offset, 256*4
+if ASIC_TARGET_ARCTURUS
+    s_mov_b32	    s_restore_accvgpr_offset_save, s_restore_accvgpr_offset
+    s_add_u32	    s_restore_accvgpr_offset, s_restore_accvgpr_offset, 256*4
+end
     s_mov_b32	    m0, 4				//VGPR initial index value = 1
     s_set_gpr_idx_on  m0, 0x8			    //M0[7:0] = M0[7:0] and M0[15:12] = 0x8
     s_add_u32	    s_restore_alloc_size, s_restore_alloc_size, 0x8000			    //add 0x8000 since we compare m0 against it later
 
   L_RESTORE_VGPR_LOOP:
-    if(USE_MTBUF_INSTEAD_OF_MUBUF)
-	tbuffer_load_format_x v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset format:BUF_NUM_FORMAT_FLOAT format: BUF_DATA_FORMAT_32 slc:1 glc:1
-    else
+
+if ASIC_TARGET_ARCTURUS
+	buffer_load_dword v0, v0, s_restore_buf_rsrc0, s_restore_accvgpr_offset slc:1 glc:1
+	buffer_load_dword v1, v0, s_restore_buf_rsrc0, s_restore_accvgpr_offset slc:1 glc:1 offset:256
+	buffer_load_dword v2, v0, s_restore_buf_rsrc0, s_restore_accvgpr_offset slc:1 glc:1 offset:256*2
+	buffer_load_dword v3, v0, s_restore_buf_rsrc0, s_restore_accvgpr_offset slc:1 glc:1 offset:256*3
+	s_add_u32 s_restore_accvgpr_offset, s_restore_accvgpr_offset, 256*4
+	s_waitcnt vmcnt(0)
+
+	for var vgpr = 0; vgpr < 4; ++ vgpr
+		v_accvgpr_write acc[vgpr], v[vgpr]
+	end
+end
+
 	buffer_load_dword v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset slc:1 glc:1
 	buffer_load_dword v1, v0, s_restore_buf_rsrc0, s_restore_mem_offset slc:1 glc:1 offset:256
 	buffer_load_dword v2, v0, s_restore_buf_rsrc0, s_restore_mem_offset slc:1 glc:1 offset:256*2
 	buffer_load_dword v3, v0, s_restore_buf_rsrc0, s_restore_mem_offset slc:1 glc:1 offset:256*3
-    end
     s_waitcnt	    vmcnt(0)								    //ensure data ready
     v_mov_b32	    v0, v0								    //v[0+m0] = v0
     v_mov_b32	    v1, v1
@@ -982,16 +781,22 @@ else
     s_cbranch_scc1  L_RESTORE_VGPR_LOOP							    //VGPR restore (except v0) is complete?
     s_set_gpr_idx_off
 											    /* VGPR restore on v0 */
-    if(USE_MTBUF_INSTEAD_OF_MUBUF)
-	tbuffer_load_format_x v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save format:BUF_NUM_FORMAT_FLOAT format: BUF_DATA_FORMAT_32 slc:1 glc:1
-    else
+if ASIC_TARGET_ARCTURUS
+	buffer_load_dword v0, v0, s_restore_buf_rsrc0, s_restore_accvgpr_offset_save slc:1 glc:1
+	buffer_load_dword v1, v0, s_restore_buf_rsrc0, s_restore_accvgpr_offset_save slc:1 glc:1 offset:256
+	buffer_load_dword v2, v0, s_restore_buf_rsrc0, s_restore_accvgpr_offset_save slc:1 glc:1 offset:256*2
+	buffer_load_dword v3, v0, s_restore_buf_rsrc0, s_restore_accvgpr_offset_save slc:1 glc:1 offset:256*3
+	s_waitcnt vmcnt(0)
+
+	for var vgpr = 0; vgpr < 4; ++ vgpr
+		v_accvgpr_write acc[vgpr], v[vgpr]
+	end
+end
+
 	buffer_load_dword v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save    slc:1 glc:1
 	buffer_load_dword v1, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save    slc:1 glc:1 offset:256
 	buffer_load_dword v2, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save    slc:1 glc:1 offset:256*2
 	buffer_load_dword v3, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save    slc:1 glc:1 offset:256*3
-    end
-
-end
 
     /*		restore SGPRs	    */
     //////////////////////////////
@@ -1007,16 +812,8 @@ end
     s_add_u32	    s_restore_alloc_size, s_restore_alloc_size, 1
     s_lshl_b32	    s_restore_alloc_size, s_restore_alloc_size, 4			    //Number of SGPRs = (sgpr_size + 1) * 16   (non-zero value)
 
-    if (SGPR_SAVE_USE_SQC)
 	s_lshl_b32	s_restore_buf_rsrc2,	s_restore_alloc_size, 2			    //NUM_RECORDS in bytes
-    else
-	s_lshl_b32	s_restore_buf_rsrc2,	s_restore_alloc_size, 8			    //NUM_RECORDS in bytes (64 threads)
-    end
-    if (SWIZZLE_EN)
-	s_add_u32	s_restore_buf_rsrc2, s_restore_buf_rsrc2, 0x0			    //FIXME need to use swizzle to enable bounds checking?
-    else
 	s_mov_b32	s_restore_buf_rsrc2,  0x1000000					    //NUM_RECORDS in bytes
-    end
 
     s_mov_b32 m0, s_restore_alloc_size
 
@@ -1044,11 +841,6 @@ end
   L_RESTORE_HWREG:
 
 
-if G8SR_DEBUG_TIMESTAMP
-      s_mov_b32 s_g8sr_ts_restore_s[0], s_restore_pc_lo
-      s_mov_b32 s_g8sr_ts_restore_s[1], s_restore_pc_hi
-end
-
     // HWREG SR memory offset : size(VGPR)+size(SGPR)
     get_vgpr_size_bytes(s_restore_mem_offset)
     get_sgpr_size_bytes(s_restore_tmp)
@@ -1056,11 +848,7 @@ end
 
 
     s_mov_b32	    s_restore_buf_rsrc2, 0x4						    //NUM_RECORDS   in bytes
-    if (SWIZZLE_EN)
-	s_add_u32	s_restore_buf_rsrc2, s_restore_buf_rsrc2, 0x0			    //FIXME need to use swizzle to enable bounds checking?
-    else
 	s_mov_b32	s_restore_buf_rsrc2,  0x1000000					    //NUM_RECORDS in bytes
-    end
 
     read_hwreg_from_mem(s_restore_m0, s_restore_buf_rsrc0, s_restore_mem_offset)		    //M0
     read_hwreg_from_mem(s_restore_pc_lo, s_restore_buf_rsrc0, s_restore_mem_offset)		//PC
@@ -1075,16 +863,6 @@ end
 
     s_waitcnt	    lgkmcnt(0)											    //from now on, it is safe to restore STATUS and IB_STS
 
-    //for normal save & restore, the saved PC points to the next inst to execute, no adjustment needs to be made, otherwise:
-    if ((EMU_RUN_HACK) && (!EMU_RUN_HACK_RESTORE_NORMAL))
-	s_add_u32 s_restore_pc_lo, s_restore_pc_lo, 8		 //pc[31:0]+8	  //two back-to-back s_trap are used (first for save and second for restore)
-	s_addc_u32  s_restore_pc_hi, s_restore_pc_hi, 0x0	 //carry bit over
-    end
-    if ((EMU_RUN_HACK) && (EMU_RUN_HACK_RESTORE_NORMAL))
-	s_add_u32 s_restore_pc_lo, s_restore_pc_lo, 4		 //pc[31:0]+4	  // save is hack through s_trap but restore is normal
-	s_addc_u32  s_restore_pc_hi, s_restore_pc_hi, 0x0	 //carry bit over
-    end
-
     s_mov_b32	    m0,		s_restore_m0
     s_mov_b32	    exec_lo,	s_restore_exec_lo
     s_mov_b32	    exec_hi,	s_restore_exec_hi
@@ -1131,11 +909,6 @@ end
 
     s_barrier							//barrier to ensure the readiness of LDS before access attempts from any other wave in the same TG //FIXME not performance-optimal at this time
 
-if G8SR_DEBUG_TIMESTAMP
-    s_memrealtime s_g8sr_ts_restore_d
-    s_waitcnt lgkmcnt(0)
-end
-
 //  s_rfe_b64 s_restore_pc_lo					//Return to the main shader program and resume execution
     s_rfe_restore_b64  s_restore_pc_lo, s_restore_m0		// s_restore_m0[0] is used to set STATUS.inst_atc
 
@@ -1190,7 +963,39 @@ function read_16sgpr_from_mem(s, s_rsrc, s_mem_offset)
     s_sub_u32	    s_mem_offset, s_mem_offset, 4*16
 end
 
+function check_if_tcp_store_ok
+	// If STATUS.ALLOW_REPLAY=0 and TRAPSTS.XNACK_ERROR=1 then TCP stores will fail.
+	s_and_b32 s_save_tmp, s_save_status, SQ_WAVE_STATUS_ALLOW_REPLAY_MASK
+	s_cbranch_scc1 L_TCP_STORE_CHECK_DONE
+
+	s_getreg_b32 s_save_tmp, hwreg(HW_REG_TRAPSTS)
+	s_andn2_b32 s_save_tmp, SQ_WAVE_TRAPSTS_XNACK_ERROR_MASK, s_save_tmp
+
+L_TCP_STORE_CHECK_DONE:
+end
+
+function write_vgpr_to_mem_with_sqc(v, s_rsrc, s_mem_offset)
+	s_mov_b32 s4, 0
 
+L_WRITE_VGPR_LANE_LOOP:
+	for var lane = 0; lane < 4; ++ lane
+		v_readlane_b32 s[lane], v, s4
+		s_add_u32 s4, s4, 1
+	end
+
+	s_buffer_store_dwordx4 s[0:3], s_rsrc, s_mem_offset glc:1
+	ack_sqc_store_workaround()
+
+	s_add_u32 s_mem_offset, s_mem_offset, 0x10
+	s_cmp_eq_u32 s4, 0x40
+	s_cbranch_scc0 L_WRITE_VGPR_LANE_LOOP
+end
+
+function write_vgprs_to_mem_with_sqc(v, n_vgprs, s_rsrc, s_mem_offset)
+	for var vgpr = 0; vgpr < n_vgprs; ++ vgpr
+		write_vgpr_to_mem_with_sqc(v[vgpr], s_rsrc, s_mem_offset)
+	end
+end
 
 function get_lds_size_bytes(s_lds_size_byte)
     // SQ LDS granularity is 64DW, while PGM_RSRC2.lds_size is in granularity 128DW
@@ -1202,6 +1007,10 @@ function get_vgpr_size_bytes(s_vgpr_size_byte)
     s_getreg_b32   s_vgpr_size_byte, hwreg(HW_REG_GPR_ALLOC,SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SHIFT,SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SIZE)	 //vpgr_size
     s_add_u32	   s_vgpr_size_byte, s_vgpr_size_byte, 1
     s_lshl_b32	   s_vgpr_size_byte, s_vgpr_size_byte, (2+8) //Number of VGPRs = (vgpr_size + 1) * 4 * 64 * 4	(non-zero value)   //FIXME for GFX, zero is possible
+
+if ASIC_TARGET_ARCTURUS
+    s_lshl_b32     s_vgpr_size_byte, s_vgpr_size_byte, 1  // Double size for ACC VGPRs
+end
 end
 
 function get_sgpr_size_bytes(s_sgpr_size_byte)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_crat.c b/drivers/gpu/drm/amd/amdkfd/kfd_crat.c
index 4e3fc284f6ac..66387caf966e 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_crat.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_crat.c
@@ -662,6 +662,7 @@ static int kfd_fill_gpu_cache_info(struct kfd_dev *kdev,
 	case CHIP_VEGA10:
 	case CHIP_VEGA12:
 	case CHIP_VEGA20:
+	case CHIP_ARCTURUS:
 		pcache_info = vega10_cache_info;
 		num_of_cache_types = ARRAY_SIZE(vega10_cache_info);
 		break;
@@ -788,7 +789,7 @@ int kfd_create_crat_image_acpi(void **crat_image, size_t *size)
  * is put in the code to ensure we don't overwrite.
  */
 #define VCRAT_SIZE_FOR_CPU	(2 * PAGE_SIZE)
-#define VCRAT_SIZE_FOR_GPU	(3 * PAGE_SIZE)
+#define VCRAT_SIZE_FOR_GPU	(4 * PAGE_SIZE)
 
 /* kfd_fill_cu_for_cpu - Fill in Compute info for the given CPU NUMA node
  *
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index 3322a443dfb2..0dc1084b5e82 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -42,6 +42,7 @@ static atomic_t kfd_locked = ATOMIC_INIT(0);
 #ifdef KFD_SUPPORT_IOMMU_V2
 static const struct kfd_device_info kaveri_device_info = {
 	.asic_family = CHIP_KAVERI,
+	.asic_name = "kaveri",
 	.max_pasid_bits = 16,
 	/* max num of queues for KV.TODO should be a dynamic value */
 	.max_no_of_hqd	= 24,
@@ -60,6 +61,7 @@ static const struct kfd_device_info kaveri_device_info = {
 
 static const struct kfd_device_info carrizo_device_info = {
 	.asic_family = CHIP_CARRIZO,
+	.asic_name = "carrizo",
 	.max_pasid_bits = 16,
 	/* max num of queues for CZ.TODO should be a dynamic value */
 	.max_no_of_hqd	= 24,
@@ -78,6 +80,7 @@ static const struct kfd_device_info carrizo_device_info = {
 
 static const struct kfd_device_info raven_device_info = {
 	.asic_family = CHIP_RAVEN,
+	.asic_name = "raven",
 	.max_pasid_bits = 16,
 	.max_no_of_hqd  = 24,
 	.doorbell_size  = 8,
@@ -96,6 +99,7 @@ static const struct kfd_device_info raven_device_info = {
 
 static const struct kfd_device_info hawaii_device_info = {
 	.asic_family = CHIP_HAWAII,
+	.asic_name = "hawaii",
 	.max_pasid_bits = 16,
 	/* max num of queues for KV.TODO should be a dynamic value */
 	.max_no_of_hqd	= 24,
@@ -114,6 +118,7 @@ static const struct kfd_device_info hawaii_device_info = {
 
 static const struct kfd_device_info tonga_device_info = {
 	.asic_family = CHIP_TONGA,
+	.asic_name = "tonga",
 	.max_pasid_bits = 16,
 	.max_no_of_hqd  = 24,
 	.doorbell_size  = 4,
@@ -131,6 +136,7 @@ static const struct kfd_device_info tonga_device_info = {
 
 static const struct kfd_device_info fiji_device_info = {
 	.asic_family = CHIP_FIJI,
+	.asic_name = "fiji",
 	.max_pasid_bits = 16,
 	.max_no_of_hqd  = 24,
 	.doorbell_size  = 4,
@@ -148,6 +154,7 @@ static const struct kfd_device_info fiji_device_info = {
 
 static const struct kfd_device_info fiji_vf_device_info = {
 	.asic_family = CHIP_FIJI,
+	.asic_name = "fiji",
 	.max_pasid_bits = 16,
 	.max_no_of_hqd  = 24,
 	.doorbell_size  = 4,
@@ -166,6 +173,7 @@ static const struct kfd_device_info fiji_vf_device_info = {
 
 static const struct kfd_device_info polaris10_device_info = {
 	.asic_family = CHIP_POLARIS10,
+	.asic_name = "polaris10",
 	.max_pasid_bits = 16,
 	.max_no_of_hqd  = 24,
 	.doorbell_size  = 4,
@@ -183,6 +191,7 @@ static const struct kfd_device_info polaris10_device_info = {
 
 static const struct kfd_device_info polaris10_vf_device_info = {
 	.asic_family = CHIP_POLARIS10,
+	.asic_name = "polaris10",
 	.max_pasid_bits = 16,
 	.max_no_of_hqd  = 24,
 	.doorbell_size  = 4,
@@ -200,6 +209,7 @@ static const struct kfd_device_info polaris10_vf_device_info = {
 
 static const struct kfd_device_info polaris11_device_info = {
 	.asic_family = CHIP_POLARIS11,
+	.asic_name = "polaris11",
 	.max_pasid_bits = 16,
 	.max_no_of_hqd  = 24,
 	.doorbell_size  = 4,
@@ -217,6 +227,7 @@ static const struct kfd_device_info polaris11_device_info = {
 
 static const struct kfd_device_info polaris12_device_info = {
 	.asic_family = CHIP_POLARIS12,
+	.asic_name = "polaris12",
 	.max_pasid_bits = 16,
 	.max_no_of_hqd  = 24,
 	.doorbell_size  = 4,
@@ -234,6 +245,7 @@ static const struct kfd_device_info polaris12_device_info = {
 
 static const struct kfd_device_info vegam_device_info = {
 	.asic_family = CHIP_VEGAM,
+	.asic_name = "vegam",
 	.max_pasid_bits = 16,
 	.max_no_of_hqd  = 24,
 	.doorbell_size  = 4,
@@ -251,6 +263,7 @@ static const struct kfd_device_info vegam_device_info = {
 
 static const struct kfd_device_info vega10_device_info = {
 	.asic_family = CHIP_VEGA10,
+	.asic_name = "vega10",
 	.max_pasid_bits = 16,
 	.max_no_of_hqd  = 24,
 	.doorbell_size  = 8,
@@ -268,6 +281,7 @@ static const struct kfd_device_info vega10_device_info = {
 
 static const struct kfd_device_info vega10_vf_device_info = {
 	.asic_family = CHIP_VEGA10,
+	.asic_name = "vega10",
 	.max_pasid_bits = 16,
 	.max_no_of_hqd  = 24,
 	.doorbell_size  = 8,
@@ -285,6 +299,7 @@ static const struct kfd_device_info vega10_vf_device_info = {
 
 static const struct kfd_device_info vega12_device_info = {
 	.asic_family = CHIP_VEGA12,
+	.asic_name = "vega12",
 	.max_pasid_bits = 16,
 	.max_no_of_hqd  = 24,
 	.doorbell_size  = 8,
@@ -302,6 +317,7 @@ static const struct kfd_device_info vega12_device_info = {
 
 static const struct kfd_device_info vega20_device_info = {
 	.asic_family = CHIP_VEGA20,
+	.asic_name = "vega20",
 	.max_pasid_bits = 16,
 	.max_no_of_hqd	= 24,
 	.doorbell_size	= 8,
@@ -317,8 +333,27 @@ static const struct kfd_device_info vega20_device_info = {
 	.num_sdma_queues_per_engine = 8,
 };
 
+static const struct kfd_device_info arcturus_device_info = {
+	.asic_family = CHIP_ARCTURUS,
+	.asic_name = "arcturus",
+	.max_pasid_bits = 16,
+	.max_no_of_hqd	= 24,
+	.doorbell_size	= 8,
+	.ih_ring_entry_size = 8 * sizeof(uint32_t),
+	.event_interrupt_class = &event_interrupt_class_v9,
+	.num_of_watch_points = 4,
+	.mqd_size_aligned = MQD_SIZE_ALIGNED,
+	.supports_cwsr = true,
+	.needs_iommu_device = false,
+	.needs_pci_atomics = false,
+	.num_sdma_engines = 2,
+	.num_xgmi_sdma_engines = 6,
+	.num_sdma_queues_per_engine = 8,
+};
+
 static const struct kfd_device_info navi10_device_info = {
 	.asic_family = CHIP_NAVI10,
+	.asic_name = "navi10",
 	.max_pasid_bits = 16,
 	.max_no_of_hqd  = 24,
 	.doorbell_size  = 8,
@@ -452,7 +487,10 @@ static const struct kfd_deviceid supported_devices[] = {
 	{ 0x66a4, &vega20_device_info },	/* Vega20 */
 	{ 0x66a7, &vega20_device_info },	/* Vega20 */
 	{ 0x66af, &vega20_device_info },	/* Vega20 */
-	/* Navi10 */
+	{ 0x738C, &arcturus_device_info },	/* Arcturus */
+	{ 0x7388, &arcturus_device_info },	/* Arcturus */
+	{ 0x738E, &arcturus_device_info },	/* Arcturus */
+	{ 0x7390, &arcturus_device_info },	/* Arcturus vf */
 	{ 0x7310, &navi10_device_info },	/* Navi10 */
 	{ 0x7312, &navi10_device_info },	/* Navi10 */
 	{ 0x7318, &navi10_device_info },	/* Navi10 */
@@ -536,6 +574,10 @@ static void kfd_cwsr_init(struct kfd_dev *kfd)
 			BUILD_BUG_ON(sizeof(cwsr_trap_gfx8_hex) > PAGE_SIZE);
 			kfd->cwsr_isa = cwsr_trap_gfx8_hex;
 			kfd->cwsr_isa_size = sizeof(cwsr_trap_gfx8_hex);
+		} else if (kfd->device_info->asic_family == CHIP_ARCTURUS) {
+			BUILD_BUG_ON(sizeof(cwsr_trap_arcturus_hex) > PAGE_SIZE);
+			kfd->cwsr_isa = cwsr_trap_arcturus_hex;
+			kfd->cwsr_isa_size = sizeof(cwsr_trap_arcturus_hex);
 		} else if (kfd->device_info->asic_family < CHIP_NAVI10) {
 			BUILD_BUG_ON(sizeof(cwsr_trap_gfx9_hex) > PAGE_SIZE);
 			kfd->cwsr_isa = cwsr_trap_gfx9_hex;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
index e6a4288bfaa6..d985e31fcc1e 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
@@ -880,8 +880,8 @@ static int initialize_nocpsch(struct device_queue_manager *dqm)
 	}
 
 	dqm->vmid_bitmap = (1 << dqm->dev->vm_info.vmid_num_kfd) - 1;
-	dqm->sdma_bitmap = (1ULL << get_num_sdma_queues(dqm)) - 1;
-	dqm->xgmi_sdma_bitmap = (1ULL << get_num_xgmi_sdma_queues(dqm)) - 1;
+	dqm->sdma_bitmap = ~0ULL >> (64 - get_num_sdma_queues(dqm));
+	dqm->xgmi_sdma_bitmap = ~0ULL >> (64 - get_num_xgmi_sdma_queues(dqm));
 
 	return 0;
 }
@@ -1019,8 +1019,8 @@ static int initialize_cpsch(struct device_queue_manager *dqm)
 	dqm->sdma_queue_count = 0;
 	dqm->xgmi_sdma_queue_count = 0;
 	dqm->active_runlist = false;
-	dqm->sdma_bitmap = (1ULL << get_num_sdma_queues(dqm)) - 1;
-	dqm->xgmi_sdma_bitmap = (1ULL << get_num_xgmi_sdma_queues(dqm)) - 1;
+	dqm->sdma_bitmap = ~0ULL >> (64 - get_num_sdma_queues(dqm));
+	dqm->xgmi_sdma_bitmap = ~0ULL >> (64 - get_num_xgmi_sdma_queues(dqm));
 
 	INIT_WORK(&dqm->hw_exception_work, kfd_process_hw_exception);
 
@@ -1786,6 +1786,7 @@ struct device_queue_manager *device_queue_manager_init(struct kfd_dev *dev)
 	case CHIP_VEGA12:
 	case CHIP_VEGA20:
 	case CHIP_RAVEN:
+	case CHIP_ARCTURUS:
 		device_queue_manager_init_v9(&dqm->asic_ops);
 		break;
 	case CHIP_NAVI10:
@@ -1813,7 +1814,8 @@ out_free:
 	return NULL;
 }
 
-void deallocate_hiq_sdma_mqd(struct kfd_dev *dev, struct kfd_mem_obj *mqd)
+static void deallocate_hiq_sdma_mqd(struct kfd_dev *dev,
+				    struct kfd_mem_obj *mqd)
 {
 	WARN(!mqd, "No hiq sdma mqd trunk to free");
 
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c b/drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c
index 60521366dd31..9dc4bff8085e 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c
@@ -405,6 +405,7 @@ int kfd_init_apertures(struct kfd_process *process)
 			case CHIP_VEGA12:
 			case CHIP_VEGA20:
 			case CHIP_RAVEN:
+			case CHIP_ARCTURUS:
 			case CHIP_NAVI10:
 				kfd_init_apertures_v9(pdd, id);
 				break;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c
index a85904ad0d5f..3ef67d2e0d9f 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c
@@ -80,6 +80,7 @@ static bool event_interrupt_isr_v9(struct kfd_dev *dev,
 		source_id == SOC15_INTSRC_SQ_INTERRUPT_MSG ||
 		source_id == SOC15_INTSRC_CP_BAD_OPCODE ||
 		client_id == SOC15_IH_CLIENTID_VMC ||
+		client_id == SOC15_IH_CLIENTID_VMC1 ||
 		client_id == SOC15_IH_CLIENTID_UTCL2;
 }
 
@@ -104,6 +105,7 @@ static void event_interrupt_wq_v9(struct kfd_dev *dev,
 	else if (source_id == SOC15_INTSRC_CP_BAD_OPCODE)
 		kfd_signal_hw_exception_event(pasid);
 	else if (client_id == SOC15_IH_CLIENTID_VMC ||
+		client_id == SOC15_IH_CLIENTID_VMC1 ||
 		 client_id == SOC15_IH_CLIENTID_UTCL2) {
 		struct kfd_vm_fault_info info = {0};
 		uint16_t ring_id = SOC15_RING_ID_FROM_IH_ENTRY(ih_ring_entry);
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c b/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c
index 29c0bd2d7a5c..8b4564f71a7a 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c
@@ -330,6 +330,7 @@ struct kernel_queue *kernel_queue_init(struct kfd_dev *dev,
 	case CHIP_VEGA12:
 	case CHIP_VEGA20:
 	case CHIP_RAVEN:
+	case CHIP_ARCTURUS:
 		kernel_queue_init_v9(&kq->ops_asic_specific);
 		break;
 	case CHIP_NAVI10:
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue_v9.c b/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue_v9.c
index 2d5ddf199bd0..9a4bafb2e175 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue_v9.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue_v9.c
@@ -81,7 +81,8 @@ static int pm_map_process_v9(struct packet_manager *pm,
 	packet->bitfields2.diq_enable = (qpd->is_debug) ? 1 : 0;
 	packet->bitfields2.process_quantum = 1;
 	packet->bitfields2.pasid = qpd->pqm->process->pasid;
-	packet->bitfields14.gds_size = qpd->gds_size;
+	packet->bitfields14.gds_size = qpd->gds_size & 0x3F;
+	packet->bitfields14.gds_size_hi = (qpd->gds_size >> 6) & 0xF;
 	packet->bitfields14.num_gws = qpd->num_gws;
 	packet->bitfields14.num_oac = qpd->num_oac;
 	packet->bitfields14.sdma_enable = 1;
@@ -143,6 +144,34 @@ static int pm_runlist_v9(struct packet_manager *pm, uint32_t *buffer,
 	return 0;
 }
 
+static int pm_set_resources_v9(struct packet_manager *pm, uint32_t *buffer,
+				struct scheduling_resources *res)
+{
+	struct pm4_mes_set_resources *packet;
+
+	packet = (struct pm4_mes_set_resources *)buffer;
+	memset(buffer, 0, sizeof(struct pm4_mes_set_resources));
+
+	packet->header.u32All = pm_build_pm4_header(IT_SET_RESOURCES,
+					sizeof(struct pm4_mes_set_resources));
+
+	packet->bitfields2.queue_type =
+			queue_type__mes_set_resources__hsa_interface_queue_hiq;
+	packet->bitfields2.vmid_mask = res->vmid_mask;
+	packet->bitfields2.unmap_latency = KFD_UNMAP_LATENCY_MS / 100;
+	packet->bitfields7.oac_mask = res->oac_mask;
+	packet->bitfields8.gds_heap_base = res->gds_heap_base;
+	packet->bitfields8.gds_heap_size = res->gds_heap_size;
+
+	packet->gws_mask_lo = lower_32_bits(res->gws_mask);
+	packet->gws_mask_hi = upper_32_bits(res->gws_mask);
+
+	packet->queue_mask_lo = lower_32_bits(res->queue_mask);
+	packet->queue_mask_hi = upper_32_bits(res->queue_mask);
+
+	return 0;
+}
+
 static int pm_map_queues_v9(struct packet_manager *pm, uint32_t *buffer,
 		struct queue *q, bool is_static)
 {
@@ -161,6 +190,8 @@ static int pm_map_queues_v9(struct packet_manager *pm, uint32_t *buffer,
 	packet->bitfields2.engine_sel =
 		engine_sel__mes_map_queues__compute_vi;
 	packet->bitfields2.gws_control_queue = q->gws ? 1 : 0;
+	packet->bitfields2.extended_engine_sel =
+		extended_engine_sel__mes_map_queues__legacy_engine_sel;
 	packet->bitfields2.queue_type =
 		queue_type__mes_map_queues__normal_compute_vi;
 
@@ -176,9 +207,15 @@ static int pm_map_queues_v9(struct packet_manager *pm, uint32_t *buffer,
 		break;
 	case KFD_QUEUE_TYPE_SDMA:
 	case KFD_QUEUE_TYPE_SDMA_XGMI:
-		packet->bitfields2.engine_sel = q->properties.sdma_engine_id +
-				engine_sel__mes_map_queues__sdma0_vi;
 		use_static = false; /* no static queues under SDMA */
+		if (q->properties.sdma_engine_id < 2)
+			packet->bitfields2.engine_sel = q->properties.sdma_engine_id +
+				engine_sel__mes_map_queues__sdma0_vi;
+		else {
+			packet->bitfields2.extended_engine_sel =
+				extended_engine_sel__mes_map_queues__sdma0_to_7_sel;
+			packet->bitfields2.engine_sel = q->properties.sdma_engine_id;
+		}
 		break;
 	default:
 		WARN(1, "queue type %d", q->properties.type);
@@ -218,13 +255,23 @@ static int pm_unmap_queues_v9(struct packet_manager *pm, uint32_t *buffer,
 	switch (type) {
 	case KFD_QUEUE_TYPE_COMPUTE:
 	case KFD_QUEUE_TYPE_DIQ:
+		packet->bitfields2.extended_engine_sel =
+			extended_engine_sel__mes_unmap_queues__legacy_engine_sel;
 		packet->bitfields2.engine_sel =
 			engine_sel__mes_unmap_queues__compute;
 		break;
 	case KFD_QUEUE_TYPE_SDMA:
 	case KFD_QUEUE_TYPE_SDMA_XGMI:
-		packet->bitfields2.engine_sel =
-			engine_sel__mes_unmap_queues__sdma0 + sdma_engine;
+		if (sdma_engine < 2) {
+			packet->bitfields2.extended_engine_sel =
+				extended_engine_sel__mes_unmap_queues__legacy_engine_sel;
+			packet->bitfields2.engine_sel =
+				engine_sel__mes_unmap_queues__sdma0 + sdma_engine;
+		} else {
+			packet->bitfields2.extended_engine_sel =
+				extended_engine_sel__mes_unmap_queues__sdma0_to_7_sel;
+			packet->bitfields2.engine_sel = sdma_engine;
+		}
 		break;
 	default:
 		WARN(1, "queue type %d", type);
@@ -326,7 +373,7 @@ static int pm_release_mem_v9(uint64_t gpu_addr, uint32_t *buffer)
 const struct packet_manager_funcs kfd_v9_pm_funcs = {
 	.map_process		= pm_map_process_v9,
 	.runlist		= pm_runlist_v9,
-	.set_resources		= pm_set_resources_vi,
+	.set_resources		= pm_set_resources_v9,
 	.map_queues		= pm_map_queues_v9,
 	.unmap_queues		= pm_unmap_queues_v9,
 	.query_status		= pm_query_status_v9,
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c
index d6cf391da591..88813dad731f 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c
@@ -98,8 +98,8 @@ void mqd_symmetrically_map_cu_mask(struct mqd_manager *mm,
 		uint32_t *se_mask)
 {
 	struct kfd_cu_info cu_info;
-	uint32_t cu_per_sh[4] = {0};
-	int i, se, cu = 0;
+	uint32_t cu_per_se[KFD_MAX_NUM_SE] = {0};
+	int i, se, sh, cu = 0;
 
 	amdgpu_amdkfd_get_cu_info(mm->dev->kgd, &cu_info);
 
@@ -107,8 +107,8 @@ void mqd_symmetrically_map_cu_mask(struct mqd_manager *mm,
 		cu_mask_count = cu_info.cu_active_number;
 
 	for (se = 0; se < cu_info.num_shader_engines; se++)
-		for (i = 0; i < 4; i++)
-			cu_per_sh[se] += hweight32(cu_info.cu_bitmap[se][i]);
+		for (sh = 0; sh < cu_info.num_shader_arrays_per_engine; sh++)
+			cu_per_se[se] += hweight32(cu_info.cu_bitmap[se % 4][sh + (se / 4)]);
 
 	/* Symmetrically map cu_mask to all SEs:
 	 * cu_mask[0] bit0 -> se_mask[0] bit0;
@@ -128,6 +128,6 @@ void mqd_symmetrically_map_cu_mask(struct mqd_manager *mm,
 				se = 0;
 				cu++;
 			}
-		} while (cu >= cu_per_sh[se] && cu < 32);
+		} while (cu >= cu_per_se[se] && cu < 32);
 	}
 }
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.h b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.h
index 550b61e81015..fbdb16418847 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.h
@@ -26,6 +26,8 @@
 
 #include "kfd_priv.h"
 
+#define KFD_MAX_NUM_SE 8
+
 /**
  * struct mqd_manager
  *
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c
index 0c58f91b3ff3..d3380c5bdbde 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c
@@ -46,7 +46,7 @@ static void update_cu_mask(struct mqd_manager *mm, void *mqd,
 			struct queue_properties *q)
 {
 	struct v9_mqd *m;
-	uint32_t se_mask[4] = {0}; /* 4 is the max # of SEs */
+	uint32_t se_mask[KFD_MAX_NUM_SE] = {0};
 
 	if (q->cu_mask_count == 0)
 		return;
@@ -59,12 +59,20 @@ static void update_cu_mask(struct mqd_manager *mm, void *mqd,
 	m->compute_static_thread_mgmt_se1 = se_mask[1];
 	m->compute_static_thread_mgmt_se2 = se_mask[2];
 	m->compute_static_thread_mgmt_se3 = se_mask[3];
+	m->compute_static_thread_mgmt_se4 = se_mask[4];
+	m->compute_static_thread_mgmt_se5 = se_mask[5];
+	m->compute_static_thread_mgmt_se6 = se_mask[6];
+	m->compute_static_thread_mgmt_se7 = se_mask[7];
 
-	pr_debug("update cu mask to %#x %#x %#x %#x\n",
+	pr_debug("update cu mask to %#x %#x %#x %#x %#x %#x %#x %#x\n",
 		m->compute_static_thread_mgmt_se0,
 		m->compute_static_thread_mgmt_se1,
 		m->compute_static_thread_mgmt_se2,
-		m->compute_static_thread_mgmt_se3);
+		m->compute_static_thread_mgmt_se3,
+		m->compute_static_thread_mgmt_se4,
+		m->compute_static_thread_mgmt_se5,
+		m->compute_static_thread_mgmt_se6,
+		m->compute_static_thread_mgmt_se7);
 }
 
 static void set_priority(struct v9_mqd *m, struct queue_properties *q)
@@ -125,6 +133,10 @@ static void init_mqd(struct mqd_manager *mm, void **mqd,
 	m->compute_static_thread_mgmt_se1 = 0xFFFFFFFF;
 	m->compute_static_thread_mgmt_se2 = 0xFFFFFFFF;
 	m->compute_static_thread_mgmt_se3 = 0xFFFFFFFF;
+	m->compute_static_thread_mgmt_se4 = 0xFFFFFFFF;
+	m->compute_static_thread_mgmt_se5 = 0xFFFFFFFF;
+	m->compute_static_thread_mgmt_se6 = 0xFFFFFFFF;
+	m->compute_static_thread_mgmt_se7 = 0xFFFFFFFF;
 
 	m->cp_hqd_persistent_state = CP_HQD_PERSISTENT_STATE__PRELOAD_REQ_MASK |
 			0x53 << CP_HQD_PERSISTENT_STATE__PRELOAD_SIZE__SHIFT;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c
index ccf6b2310316..2c8624c5b42c 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c
@@ -239,6 +239,7 @@ int pm_init(struct packet_manager *pm, struct device_queue_manager *dqm)
 	case CHIP_VEGA12:
 	case CHIP_VEGA20:
 	case CHIP_RAVEN:
+	case CHIP_ARCTURUS:
 		pm->pmf = &kfd_v9_pm_funcs;
 		break;
 	case CHIP_NAVI10:
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_pm4_headers_ai.h b/drivers/gpu/drm/amd/amdkfd/kfd_pm4_headers_ai.h
index e3e21404cfa0..4d7add843746 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_pm4_headers_ai.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_pm4_headers_ai.h
@@ -83,10 +83,10 @@ struct pm4_mes_set_resources {
 
 	union {
 		struct {
-		uint32_t gds_heap_base:6;
-		uint32_t reserved3:5;
-		uint32_t gds_heap_size:6;
-		uint32_t reserved4:15;
+		uint32_t gds_heap_base:10;
+		uint32_t reserved3:1;
+		uint32_t gds_heap_size:10;
+		uint32_t reserved4:11;
 		} bitfields8;
 		uint32_t ordinal8;
 	};
@@ -179,7 +179,7 @@ struct pm4_mes_map_process {
 			uint32_t num_gws:7;
 			uint32_t sdma_enable:1;
 			uint32_t num_oac:4;
-			uint32_t reserved8:4;
+			uint32_t gds_size_hi:4;
 			uint32_t gds_size:6;
 			uint32_t num_queues:10;
 		} bitfields14;
@@ -260,6 +260,10 @@ enum mes_map_queues_engine_sel_enum {
 	engine_sel__mes_map_queues__sdma1_vi = 3
 };
 
+enum mes_map_queues_extended_engine_sel_enum {
+	extended_engine_sel__mes_map_queues__legacy_engine_sel = 0,
+	extended_engine_sel__mes_map_queues__sdma0_to_7_sel = 1
+};
 
 struct pm4_mes_map_queues {
 	union {
@@ -269,7 +273,8 @@ struct pm4_mes_map_queues {
 
 	union {
 		struct {
-			uint32_t reserved1:4;
+			uint32_t reserved1:2;
+			enum mes_map_queues_extended_engine_sel_enum extended_engine_sel:2;
 			enum mes_map_queues_queue_sel_enum queue_sel:2;
 			uint32_t reserved5:6;
 			uint32_t gws_control_queue:1;
@@ -382,6 +387,11 @@ enum mes_unmap_queues_engine_sel_enum {
 	engine_sel__mes_unmap_queues__sdmal = 3
 };
 
+enum mes_unmap_queues_extended_engine_sel_enum {
+	extended_engine_sel__mes_unmap_queues__legacy_engine_sel = 0,
+	extended_engine_sel__mes_unmap_queues__sdma0_to_7_sel = 1
+};
+
 struct pm4_mes_unmap_queues {
 	union {
 		union PM4_MES_TYPE_3_HEADER   header;            /* header */
@@ -391,7 +401,7 @@ struct pm4_mes_unmap_queues {
 	union {
 		struct {
 			enum mes_unmap_queues_action_enum action:2;
-			uint32_t reserved1:2;
+			enum mes_unmap_queues_extended_engine_sel_enum extended_engine_sel:2;
 			enum mes_unmap_queues_queue_sel_enum queue_sel:2;
 			uint32_t reserved2:20;
 			enum mes_unmap_queues_engine_sel_enum engine_sel:3;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
index 3933fb6a371e..3bb75d11a662 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
@@ -195,6 +195,7 @@ struct kfd_event_interrupt_class {
 
 struct kfd_device_info {
 	enum amd_asic_type asic_family;
+	const char *asic_name;
 	const struct kfd_event_interrupt_class *event_interrupt_class;
 	unsigned int max_pasid_bits;
 	unsigned int max_no_of_hqd;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
index 8f1076c0c88a..0c6ac043ae3c 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
@@ -801,6 +801,8 @@ int kfd_process_device_init_vm(struct kfd_process_device *pdd,
 		return ret;
 	}
 
+	amdgpu_vm_set_task_info(pdd->vm);
+
 	ret = kfd_process_device_reserve_ib_mem(pdd);
 	if (ret)
 		goto err_reserve_ib_mem;
@@ -1042,7 +1044,6 @@ static void restore_process_worker(struct work_struct *work)
 {
 	struct delayed_work *dwork;
 	struct kfd_process *p;
-	struct kfd_process_device *pdd;
 	int ret = 0;
 
 	dwork = to_delayed_work(work);
@@ -1051,16 +1052,6 @@ static void restore_process_worker(struct work_struct *work)
 	 * lifetime of this thread, kfd_process p will be valid
 	 */
 	p = container_of(dwork, struct kfd_process, restore_work);
-
-	/* Call restore_process_bos on the first KGD device. This function
-	 * takes care of restoring the whole process including other devices.
-	 * Restore can fail if enough memory is not available. If so,
-	 * reschedule again.
-	 */
-	pdd = list_first_entry(&p->per_device_data,
-			       struct kfd_process_device,
-			       per_device_list);
-
 	pr_debug("Started restoring pasid %d\n", p->pasid);
 
 	/* Setting last_restore_timestamp before successful restoration.
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
index c2e6e47abaf2..7551761f2aa9 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
@@ -406,8 +406,6 @@ static ssize_t node_show(struct kobject *kobj, struct attribute *attr,
 		char *buffer)
 {
 	struct kfd_topology_device *dev;
-	char public_name[KFD_TOPOLOGY_PUBLIC_NAME_SIZE];
-	uint32_t i;
 	uint32_t log_max_watch_addr;
 
 	/* Making sure that the buffer is an empty string */
@@ -422,14 +420,8 @@ static ssize_t node_show(struct kobject *kobj, struct attribute *attr,
 	if (strcmp(attr->name, "name") == 0) {
 		dev = container_of(attr, struct kfd_topology_device,
 				attr_name);
-		for (i = 0; i < KFD_TOPOLOGY_PUBLIC_NAME_SIZE; i++) {
-			public_name[i] =
-					(char)dev->node_props.marketing_name[i];
-			if (dev->node_props.marketing_name[i] == 0)
-				break;
-		}
-		public_name[KFD_TOPOLOGY_PUBLIC_NAME_SIZE-1] = 0x0;
-		return sysfs_show_str_val(buffer, public_name);
+
+		return sysfs_show_str_val(buffer, dev->node_props.name);
 	}
 
 	dev = container_of(attr, struct kfd_topology_device,
@@ -1274,6 +1266,10 @@ int kfd_topology_add_device(struct kfd_dev *gpu)
 	 */
 
 	amdgpu_amdkfd_get_cu_info(dev->gpu->kgd, &cu_info);
+
+	strncpy(dev->node_props.name, gpu->device_info->asic_name,
+			KFD_TOPOLOGY_PUBLIC_NAME_SIZE);
+
 	dev->node_props.simd_arrays_per_engine =
 		cu_info.num_shader_arrays_per_engine;
 
@@ -1321,6 +1317,7 @@ int kfd_topology_add_device(struct kfd_dev *gpu)
 	case CHIP_VEGA12:
 	case CHIP_VEGA20:
 	case CHIP_RAVEN:
+	case CHIP_ARCTURUS:
 	case CHIP_NAVI10:
 		dev->node_props.capability |= ((HSA_CAP_DOORBELL_TYPE_2_0 <<
 			HSA_CAP_DOORBELL_TYPE_TOTALBITS_SHIFT) &
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_topology.h b/drivers/gpu/drm/amd/amdkfd/kfd_topology.h
index 276354aa0fcc..d4718d58d0f2 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_topology.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_topology.h
@@ -27,7 +27,7 @@
 #include <linux/list.h>
 #include "kfd_crat.h"
 
-#define KFD_TOPOLOGY_PUBLIC_NAME_SIZE 128
+#define KFD_TOPOLOGY_PUBLIC_NAME_SIZE 32
 
 #define HSA_CAP_HOT_PLUGGABLE			0x00000001
 #define HSA_CAP_ATS_PRESENT			0x00000002
@@ -81,7 +81,7 @@ struct kfd_node_properties {
 	int32_t  drm_render_minor;
 	uint32_t num_sdma_engines;
 	uint32_t num_sdma_xgmi_engines;
-	uint16_t marketing_name[KFD_TOPOLOGY_PUBLIC_NAME_SIZE];
+	char name[KFD_TOPOLOGY_PUBLIC_NAME_SIZE];
 };
 
 #define HSA_MEM_HEAP_TYPE_SYSTEM	0