VALU instructions with SGPR source need wait states to avoid hazard with later instructions. Signed-off-by: Jay Cornwall <jay.cornwall@xxxxxxx> Cc: Lancelot Six <lancelot.six@xxxxxxx> --- .../gpu/drm/amd/amdkfd/cwsr_trap_handler.h | 406 ++++++++++++++++-- .../amd/amdkfd/cwsr_trap_handler_gfx12.asm | 13 + 2 files changed, 392 insertions(+), 27 deletions(-) diff --git a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h index 651660958e5b..112f28ff0018 100644 --- a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h +++ b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h @@ -3644,7 +3644,7 @@ static const uint32_t cwsr_trap_gfx9_4_3_hex[] = { }; static const uint32_t cwsr_trap_gfx12_hex[] = { - 0xbfa00001, 0xbfa0024b, + 0xbfa00001, 0xbfa0050b, 0xb0804009, 0xb8f8f804, 0x9178ff78, 0x00008c00, 0xb8fbf811, 0x8b6eff78, @@ -3718,7 +3718,15 @@ static const uint32_t cwsr_trap_gfx12_hex[] = { 0x00011677, 0xd7610000, 0x00011a79, 0xd7610000, 0x00011c7e, 0xd7610000, - 0x00011e7f, 0xbefe00ff, + 0x00011e7f, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xbefe00ff, 0x00003fff, 0xbeff0080, 0xee0a407a, 0x000c0000, 0x00004000, 0xd760007a, @@ -3759,33 +3767,153 @@ static const uint32_t cwsr_trap_gfx12_hex[] = { 0xbf94fffe, 0xb8faf804, 0x8b7a847a, 0x91788478, 0x8c787a78, 0xd7610002, - 0x0000fa71, 0x807d817d, + 0x0000fa71, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0x807d817d, 0xd7610002, 0x0000fa6c, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, 0x807d817d, 0x917aff6d, 0x80000000, 0xd7610002, - 0x0000fa7a, 0x807d817d, + 0x0000fa7a, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0x807d817d, 0xd7610002, 0x0000fa6e, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, 0x807d817d, 0xd7610002, - 0x0000fa6f, 0x807d817d, + 0x0000fa6f, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0x807d817d, 0xd7610002, 0x0000fa78, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, 0x807d817d, 0xb8faf811, 0xd7610002, 0x0000fa7a, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, 0x807d817d, 0xd7610002, - 0x0000fa7b, 0x807d817d, + 0x0000fa7b, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0x807d817d, 0xb8f1f801, 0xd7610002, - 0x0000fa71, 0x807d817d, + 0x0000fa71, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0x807d817d, 0xb8f1f814, 0xd7610002, - 0x0000fa71, 0x807d817d, + 0x0000fa71, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0x807d817d, 0xb8f1f815, 0xd7610002, - 0x0000fa71, 0x807d817d, + 0x0000fa71, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0x807d817d, 0xb8f1f812, 0xd7610002, - 0x0000fa71, 0x807d817d, + 0x0000fa71, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0x807d817d, 0xb8f1f813, 0xd7610002, - 0x0000fa71, 0x807d817d, + 0x0000fa71, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0x807d817d, 0xb8faf802, 0xd7610002, - 0x0000fa7a, 0x807d817d, + 0x0000fa7a, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0x807d817d, 0xbefa50c1, 0xbfc70000, 0xd7610002, 0x0000fa7a, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, 0x807d817d, 0xbefe00ff, 0x0000ffff, 0xbeff0080, 0xc4068070, 0x008ce802, @@ -3802,57 +3930,281 @@ static const uint32_t cwsr_trap_gfx12_hex[] = { 0xbe864106, 0xbe884108, 0xbe8a410a, 0xbe8c410c, 0xbe8e410e, 0xd7610002, - 0x0000f200, 0x80798179, + 0x0000f200, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0x80798179, 0xd7610002, 0x0000f201, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, 0x80798179, 0xd7610002, - 0x0000f202, 0x80798179, + 0x0000f202, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0x80798179, 0xd7610002, 0x0000f203, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, 0x80798179, 0xd7610002, - 0x0000f204, 0x80798179, + 0x0000f204, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0x80798179, 0xd7610002, 0x0000f205, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, 0x80798179, 0xd7610002, - 0x0000f206, 0x80798179, + 0x0000f206, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0x80798179, 0xd7610002, 0x0000f207, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, 0x80798179, 0xd7610002, - 0x0000f208, 0x80798179, + 0x0000f208, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0x80798179, 0xd7610002, 0x0000f209, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, 0x80798179, 0xd7610002, - 0x0000f20a, 0x80798179, + 0x0000f20a, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0x80798179, 0xd7610002, 0x0000f20b, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, 0x80798179, 0xd7610002, - 0x0000f20c, 0x80798179, + 0x0000f20c, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0x80798179, 0xd7610002, 0x0000f20d, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, 0x80798179, 0xd7610002, - 0x0000f20e, 0x80798179, + 0x0000f20e, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0x80798179, 0xd7610002, 0x0000f20f, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, 0x80798179, 0xbf06a079, 0xbfa10007, 0xc4068070, 0x008ce802, 0x00000000, 0x8070ff70, 0x00000080, 0xbef90080, 0x7e040280, 0x807d907d, 0xbf0aff7d, - 0x00000060, 0xbfa2ffbb, + 0x00000060, 0xbfa2febb, 0xbe804100, 0xbe824102, 0xbe844104, 0xbe864106, 0xbe884108, 0xbe8a410a, 0xd7610002, 0x0000f200, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, 0x80798179, 0xd7610002, - 0x0000f201, 0x80798179, + 0x0000f201, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0x80798179, 0xd7610002, 0x0000f202, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, 0x80798179, 0xd7610002, - 0x0000f203, 0x80798179, + 0x0000f203, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0x80798179, 0xd7610002, 0x0000f204, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, 0x80798179, 0xd7610002, - 0x0000f205, 0x80798179, + 0x0000f205, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0x80798179, 0xd7610002, 0x0000f206, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, 0x80798179, 0xd7610002, - 0x0000f207, 0x80798179, + 0x0000f207, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0x80798179, 0xd7610002, 0x0000f208, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, 0x80798179, 0xd7610002, - 0x0000f209, 0x80798179, + 0x0000f209, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0x80798179, 0xd7610002, 0x0000f20a, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, + 0xd8500000, 0x00000000, 0x80798179, 0xd7610002, - 0x0000f20b, 0x80798179, + 0x0000f20b, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0xd8500000, + 0x00000000, 0x80798179, 0xc4068070, 0x008ce802, 0x00000000, 0xbefe00c1, 0x857d9973, 0x8b7d817d, diff --git a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx12.asm b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx12.asm index 7b9d36e5fa43..a952d4d4eeaa 100644 --- a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx12.asm +++ b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx12.asm @@ -30,6 +30,7 @@ #define CHIP_GFX12 37 #define SINGLE_STEP_MISSED_WORKAROUND 1 //workaround for lost TRAP_AFTER_INST exception when SAVECTX raised +#define HAVE_VALU_SGPR_HAZARD (ASIC_FAMILY == CHIP_GFX12) var SQ_WAVE_STATE_PRIV_BARRIER_COMPLETE_MASK = 0x4 var SQ_WAVE_STATE_PRIV_SCC_SHIFT = 9 @@ -351,6 +352,7 @@ L_HAVE_VGPRS: v_writelane_b32 v0, ttmp13, 0xD v_writelane_b32 v0, exec_lo, 0xE v_writelane_b32 v0, exec_hi, 0xF + valu_sgpr_hazard() s_mov_b32 exec_lo, 0x3FFF s_mov_b32 exec_hi, 0x0 @@ -1059,6 +1061,7 @@ end function write_hwreg_to_v2(s) // Copy into VGPR for later TCP store. v_writelane_b32 v2, s, m0 + valu_sgpr_hazard() s_add_u32 m0, m0, 0x1 end @@ -1067,6 +1070,7 @@ function write_16sgpr_to_v2(s) // Copy into VGPR for later TCP store. for var sgpr_idx = 0; sgpr_idx < 16; sgpr_idx ++ v_writelane_b32 v2, s[sgpr_idx], ttmp13 + valu_sgpr_hazard() s_add_u32 ttmp13, ttmp13, 0x1 end end @@ -1075,6 +1079,7 @@ function write_12sgpr_to_v2(s) // Copy into VGPR for later TCP store. for var sgpr_idx = 0; sgpr_idx < 12; sgpr_idx ++ v_writelane_b32 v2, s[sgpr_idx], ttmp13 + valu_sgpr_hazard() s_add_u32 ttmp13, ttmp13, 0x1 end end @@ -1128,3 +1133,11 @@ function get_wave_size2(s_reg) s_getreg_b32 s_reg, hwreg(HW_REG_WAVE_STATUS,SQ_WAVE_STATUS_WAVE64_SHIFT,SQ_WAVE_STATUS_WAVE64_SIZE) s_lshl_b32 s_reg, s_reg, S_WAVE_SIZE end + +function valu_sgpr_hazard +#if HAVE_VALU_SGPR_HAZARD + for var rep = 0; rep < 8; rep ++ + ds_nop + end +#endif +end -- 2.34.1