Commit 37f86a9b authored by Jay Cornwall's avatar Jay Cornwall Committed by Alex Deucher
Browse files

drm/amdkfd: Merge gfx9/arcturus trap handlers, add ACC VGPR save



ACC VGPRs are a secondary VGPR set of same size as the primary VGPRs.
Save them as a block immediately following VGPRs.

Signed-off-by: default avatarJay Cornwall <Jay.Cornwall@amd.com>
Signed-off-by: default avatarAlex Deucher <alexander.deucher@amd.com>
parent 5ddd4a9a
Loading
Loading
Loading
Loading
+457 −170
Original line number Diff line number Diff line
@@ -861,200 +861,487 @@ static const uint32_t cwsr_trap_gfx10_hex[] = {
	0xbf9f0000, 0x00000000,
};
static const uint32_t cwsr_trap_arcturus_hex[] = {
	0xbf820001, 0xbf8200ca,
	0xb8f0f802, 0x89708670,
	0xb8f1f803, 0x8671ff71,
	0x00000400, 0xbf850008,
	0xb8f1f803, 0x8671ff71,
	0x000001ff, 0xbf850001,
	0x806c846c, 0x866dff6d,
	0x0000ffff, 0xbe801f6c,
	0xb8f1f803, 0x8671ff71,
	0x00000100, 0xbf840006,
	0xbef60080, 0xb9760203,
	0x866dff6d, 0x0000ffff,
	0xbf820001, 0xbf8202bd,
	0xb8f8f802, 0x89788678,
	0xb8fbf803, 0x866eff7b,
	0x00000400, 0xbf85003b,
	0x866eff7b, 0x00000800,
	0xbf850003, 0x866eff7b,
	0x00000100, 0xbf84000c,
	0x866eff78, 0x00002000,
	0xbf840005, 0xbf8e0010,
	0xb8eef803, 0x866eff6e,
	0x00000400, 0xbf84fffb,
	0x8778ff78, 0x00002000,
	0x80ec886c, 0x82ed806d,
	0xbef60080, 0xb9760283,
	0xbef20068, 0xbef30069,
	0xb8f62407, 0x8e769b76,
	0x876d766d, 0xb8f603c7,
	0x8e769a76, 0x876d766d,
	0xb8f6f807, 0x8676ff76,
	0x00007fff, 0xb976f807,
	0xbeee007e, 0xbeef007f,
	0xbefe0180, 0xbf900004,
	0xbf8e0002, 0xbf88fffe,
	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,
	0x8676ff7f, 0x08000000,
	0x8f768376, 0x87777677,
	0x8676ff7f, 0x70000000,
	0x8f768176, 0x87777677,
	0xbefb007c, 0xbefa0080,
	0xbf8a0000, 0x8676ff7f,
	0x04000000, 0xbf840012,
	0xbefe00c1, 0xbeff00c1,
	0xb8f14306, 0x8671c171,
	0xbf84000d, 0x8e718671,
	0x8e718271, 0xbef60071,
	0xbef600ff, 0x01000000,
	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,
	0x807cff7c, 0x00000100,
	0x807aff7a, 0x00000100,
	0xbf0a717c, 0xbf85fffa,
	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,
	0xb8f12a05, 0x80718171,
	0x8e718271, 0x8e768871,
	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,
	0x8071ff71, 0x00001000,
	0x7e000300, 0xe0724000,
	0x7a1d0000, 0x807c817c,
	0x807aff7a, 0x00000100,
	0xbf0a717c, 0xbf85fff8,
	0xbf9c0000, 0xbefe00c1,
	0xbeff00c1, 0xb8f12a05,
	0x80718171, 0x8e718271,
	0x8e768871, 0xbef600ff,
	0x01000000, 0xbefc0080,
	0xbf11017c, 0x8071ff71,
	0x00001000, 0xd3d84000,
	0x18000100, 0x7e000000,
	0x7e000000, 0xe0724000,
	0x7a1d0000, 0x807c817c,
	0x807aff7a, 0x00000100,
	0xbf0a717c, 0xbf85fff5,
	0xbf9c0000, 0xb8f11605,
	0x80718171, 0x8e718471,
	0x8e768871, 0xbef600ff,
	0x01000000, 0xbefc0080,
	0xbf800000, 0xbe802a00,
	0x7e000200, 0xe0724000,
	0x7a1d0000, 0x807aff7a,
	0x00000100, 0x807c817c,
	0xbf0a717c, 0xbf85fff7,
	0xbef60084, 0xbef600ff,
	0x01000000, 0x7e00027b,
	0xe0724000, 0x7a1d0000,
	0x807aff7a, 0x00000100,
	0x7e00026c, 0xe0724000,
	0x7a1d0000, 0x807aff7a,
	0x00000100, 0x7e00026d,
	0xe0724000, 0x7a1d0000,
	0x807aff7a, 0x00000100,
	0x7e00026e, 0xe0724000,
	0x7a1d0000, 0x807aff7a,
	0x00000100, 0x7e00026f,
	0xe0724000, 0x7a1d0000,
	0x807aff7a, 0x00000100,
	0x7e000270, 0xe0724000,
	0x7a1d0000, 0x807aff7a,
	0x00000100, 0xb8f1f803,
	0x7e000271, 0xe0724000,
	0x7a1d0000, 0x807aff7a,
	0x00000100, 0x7e000272,
	0xe0724000, 0x7a1d0000,
	0x807aff7a, 0x00000100,
	0x7e000273, 0xe0724000,
	0x7a1d0000, 0x807aff7a,
	0x00000100, 0xb8fbf801,
	0x7e00027b, 0xe0724000,
	0x7a1d0000, 0x807aff7a,
	0x00000100, 0xbf8200bb,
	0xbef4007e, 0x8675ff7f,
	0x0000ffff, 0x8775ff75,
	0x00040000, 0xbef60080,
	0xbef700ff, 0x00807fac,
	0x8672ff7f, 0x08000000,
	0x8f728372, 0x87777277,
	0x8672ff7f, 0x70000000,
	0x8f728172, 0x87777277,
	0xbef80080, 0x8672ff7f,
	0x04000000, 0xbf840011,
	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,
	0xbf84000c, 0x8e6f866f,
	0xbf84001a, 0x8e6f866f,
	0x8e6f826f, 0xbef6006f,
	0xb8f82a05, 0x80788178,
	0x8e788a78, 0x8e788178,
	0xb8ee1605, 0x806e816e,
	0x8e6e866e, 0x80786e78,
	0x8078ff78, 0x00000080,
	0xbef600ff, 0x01000000,
	0xbefc0080, 0x807cff7c,
	0x00000100, 0x8078ff78,
	0x00000100, 0xbf0a6f7c,
	0xbf85fffa, 0xbefe00c1,
	0xbeff00c1, 0xb8ef2a05,
	0x806f816f, 0x8e6f826f,
	0x8e76886f, 0xbef600ff,
	0x01000000, 0xbef20078,
	0x8078ff78, 0x00000100,
	0xbefc0081, 0xbf11087c,
	0x806fff6f, 0x00008000,
	0xe0524000, 0x781d0000,
	0xbf8c0f70, 0x7e000300,
	0x807c817c, 0x8078ff78,
	0x00000100, 0xbf0a6f7c,
	0xbf85fff7, 0xbf9c0000,
	0xbefc0080, 0xe0510000,
	0x781d0000, 0xe0510100,
	0x781d0000, 0x807cff7c,
	0x00000200, 0x8078ff78,
	0x00000200, 0xbf0a6f7c,
	0xbf85fff6, 0xbef80080,
	0xbefe00c1, 0xbeff00c1,
	0xb8ef2a05, 0x806f816f,
	0x8e6f826f, 0x8e76886f,
	0xbef600ff, 0x01000000,
	0xbefc0080, 0xbf11087c,
	0x806fff6f, 0x00008000,
	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,
	0xbf8c0f70, 0xd3d94000,
	0x18000100, 0x807c817c,
	0x8078ff78, 0x00000100,
	0xbf0a6f7c, 0xbf85fff6,
	0xe0524100, 0x781d0100,
	0xe0524200, 0x781d0200,
	0xe0524300, 0x781d0300,
	0xbf8c0f70, 0x7e000300,
	0x7e020301, 0x7e040302,
	0x7e060303, 0x807c847c,
	0x8078ff78, 0x00000400,
	0xbf0a6f7c, 0xbf85ffdb,
	0xbf9c0000, 0xe0524000,
	0x721d0000, 0xb8ef1605,
	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,
	0x8e76886f, 0xbef600ff,
	0x01000000, 0xc0211cba,
	0x00000078, 0x8078ff78,
	0x00000100, 0xbefc0081,
	0xc021003a, 0x00000078,
	0x8078ff78, 0x00000100,
	0xbf8cc07f, 0xbe802c00,
	0xbf800000, 0x807c817c,
	0xbf0a6f7c, 0xbf85fff6,
	0xbe800072, 0xbef60084,
	0xbef600ff, 0x01000000,
	0xc0211bfa, 0x00000078,
	0x8078ff78, 0x00000100,
	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,
	0x8078ff78, 0x00000100,
	0xc0211b7a, 0x00000078,
	0x8078ff78, 0x00000100,
	0xc0211eba, 0x00000078,
	0x8078ff78, 0x00000100,
	0xc0211efa, 0x00000078,
	0x8078ff78, 0x00000100,
	0x80788478, 0xc0211b7a,
	0x00000078, 0x80788478,
	0xc0211c3a, 0x00000078,
	0x8078ff78, 0x00000100,
	0xc0211c7a, 0x00000078,
	0x8078ff78, 0x00000100,
	0x80788478, 0xc0211c7a,
	0x00000078, 0x80788478,
	0xc0211eba, 0x00000078,
	0x80788478, 0xc0211efa,
	0x00000078, 0x80788478,
	0xc0211a3a, 0x00000078,
	0x8078ff78, 0x00000100,
	0xc0211a7a, 0x00000078,
	0x8078ff78, 0x00000100,
	0x80788478, 0xc0211a7a,
	0x00000078, 0x80788478,
	0xc0211cfa, 0x00000078,
	0x8078ff78, 0x00000100,
	0xbf8cc07f, 0xbef2006d,
	0x866dff72, 0x0000ffff,
	0xbefc006f, 0xbefe007a,
	0xbeff007b, 0x866f71ff,
	0x80788478, 0xbf8cc07f,
	0xbefc006f, 0xbefe0070,
	0xbeff0071, 0x866f7bff,
	0x000003ff, 0xb96f4803,
	0x866f71ff, 0xfffff800,
	0x866f7bff, 0xfffff800,
	0x8f6f8b6f, 0xb96fa2c3,
	0xb973f801, 0x866fff72,
	0xf8000000, 0x8f6f9b6f,
	0x8e6f906f, 0xbef30080,
	0x87736f73, 0x866fff72,
	0x04000000, 0x8f6f9a6f,
	0x8e6f8f6f, 0x87736f73,
	0x866fff70, 0x00800000,
	0x8f6f976f, 0xb973f807,
	0x86fe7e7e, 0x86ea6a6a,
	0xb970f802, 0xbf8a0000,
	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,
};
+0 −746

File deleted.

Preview size limit exceeded, changes collapsed.

+81 −2
Original line number Diff line number Diff line
@@ -197,13 +197,15 @@ 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
@@ -226,7 +228,7 @@ var s_restore_ttmps_hi = s_restore_alloc_size //no conflict
/* Shader Main*/

shader main
  asic(GFX9)
  asic(DEFAULT)
  type(CS)


@@ -791,10 +793,48 @@ 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

L_SAVE_ACCVGPR_LOOP:
    for var vgpr = 0; vgpr < 4; ++ vgpr
        v_accvgpr_read v[vgpr], acc[vgpr]  // v[N] = acc[N+m0]
    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_PGM_END_SAVED  */				    //FIXME  graphics ONLY
    if ((EMU_RUN_HACK) && (!EMU_RUN_HACK_SAVE_NORMAL_EXIT))
@@ -921,6 +961,11 @@ 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 ASIC_TARGET_ARCTURUS
    s_mov_b32	    s_restore_accvgpr_offset, s_restore_buf_rsrc2                           //ACC VGPRs at end of VGPRs
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
@@ -958,6 +1003,10 @@ 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
@@ -966,6 +1015,20 @@ else
    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
@@ -982,6 +1045,18 @@ else
    s_cbranch_scc1  L_RESTORE_VGPR_LOOP							    //VGPR restore (except v0) is complete?
    s_set_gpr_idx_off
											    /* VGPR restore on v0 */
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

    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
@@ -1202,6 +1277,10 @@ 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)