From baf1d45ec29a4ddc34410087527ad8b841ec082a Mon Sep 17 00:00:00 2001 From: Tom Stellard Date: Tue, 22 Apr 2014 16:33:57 +0000 Subject: [PATCH] R600/SI: Reorganize SIInstructions.td git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206902 91177308-0d34-0410-b5e6-96231b3b80d8 --- lib/Target/R600/SIInstructions.td | 693 ++++++++++++++++-------------- 1 file changed, 381 insertions(+), 312 deletions(-) diff --git a/lib/Target/R600/SIInstructions.td b/lib/Target/R600/SIInstructions.td index a1d2cf43f78..80897f2f416 100644 --- a/lib/Target/R600/SIInstructions.td +++ b/lib/Target/R600/SIInstructions.td @@ -36,6 +36,50 @@ def WAIT_FLAG : InstFlag<"printWaitFlag">; let Predicates = [isSI] in { +//===----------------------------------------------------------------------===// +// SMRD Instructions +//===----------------------------------------------------------------------===// + +let mayLoad = 1 in { + +// We are using the SGPR_32 and not the SReg_32 register class for 32-bit +// SMRD instructions, because the SGPR_32 register class does not include M0 +// and writing to M0 from an SMRD instruction will hang the GPU. +defm S_LOAD_DWORD : SMRD_Helper <0x00, "S_LOAD_DWORD", SReg_64, SGPR_32>; +defm S_LOAD_DWORDX2 : SMRD_Helper <0x01, "S_LOAD_DWORDX2", SReg_64, SReg_64>; +defm S_LOAD_DWORDX4 : SMRD_Helper <0x02, "S_LOAD_DWORDX4", SReg_64, SReg_128>; +defm S_LOAD_DWORDX8 : SMRD_Helper <0x03, "S_LOAD_DWORDX8", SReg_64, SReg_256>; +defm S_LOAD_DWORDX16 : SMRD_Helper <0x04, "S_LOAD_DWORDX16", SReg_64, SReg_512>; + +defm S_BUFFER_LOAD_DWORD : SMRD_Helper < + 0x08, "S_BUFFER_LOAD_DWORD", SReg_128, SGPR_32 +>; + +defm S_BUFFER_LOAD_DWORDX2 : SMRD_Helper < + 0x09, "S_BUFFER_LOAD_DWORDX2", SReg_128, SReg_64 +>; + +defm S_BUFFER_LOAD_DWORDX4 : SMRD_Helper < + 0x0a, "S_BUFFER_LOAD_DWORDX4", SReg_128, SReg_128 +>; + +defm S_BUFFER_LOAD_DWORDX8 : SMRD_Helper < + 0x0b, "S_BUFFER_LOAD_DWORDX8", SReg_128, SReg_256 +>; + +defm S_BUFFER_LOAD_DWORDX16 : SMRD_Helper < + 0x0c, "S_BUFFER_LOAD_DWORDX16", SReg_128, SReg_512 +>; + +} // mayLoad = 1 + +//def S_MEMTIME : SMRD_ <0x0000001e, "S_MEMTIME", []>; +//def S_DCACHE_INV : SMRD_ <0x0000001f, "S_DCACHE_INV", []>; + +//===----------------------------------------------------------------------===// +// SOP1 Instructions +//===----------------------------------------------------------------------===// + let neverHasSideEffects = 1 in { let isMoveImm = 1 in { @@ -107,6 +151,160 @@ def S_MOVRELD_B64 : SOP1_64 <0x00000031, "S_MOVRELD_B64", []>; def S_MOV_REGRD_B32 : SOP1_32 <0x00000033, "S_MOV_REGRD_B32", []>; def S_ABS_I32 : SOP1_32 <0x00000034, "S_ABS_I32", []>; def S_MOV_FED_B32 : SOP1_32 <0x00000035, "S_MOV_FED_B32", []>; + +//===----------------------------------------------------------------------===// +// SOP2 Instructions +//===----------------------------------------------------------------------===// + +let Defs = [SCC] in { // Carry out goes to SCC +let isCommutable = 1 in { +def S_ADD_U32 : SOP2_32 <0x00000000, "S_ADD_U32", []>; +def S_ADD_I32 : SOP2_32 <0x00000002, "S_ADD_I32", + [(set i32:$dst, (add SSrc_32:$src0, SSrc_32:$src1))] +>; +} // End isCommutable = 1 + +def S_SUB_U32 : SOP2_32 <0x00000001, "S_SUB_U32", []>; +def S_SUB_I32 : SOP2_32 <0x00000003, "S_SUB_I32", + [(set i32:$dst, (sub SSrc_32:$src0, SSrc_32:$src1))] +>; + +let Uses = [SCC] in { // Carry in comes from SCC +let isCommutable = 1 in { +def S_ADDC_U32 : SOP2_32 <0x00000004, "S_ADDC_U32", + [(set i32:$dst, (adde (i32 SSrc_32:$src0), (i32 SSrc_32:$src1)))]>; +} // End isCommutable = 1 + +def S_SUBB_U32 : SOP2_32 <0x00000005, "S_SUBB_U32", + [(set i32:$dst, (sube (i32 SSrc_32:$src0), (i32 SSrc_32:$src1)))]>; +} // End Uses = [SCC] +} // End Defs = [SCC] + +def S_MIN_I32 : SOP2_32 <0x00000006, "S_MIN_I32", + [(set i32:$dst, (AMDGPUsmin i32:$src0, i32:$src1))] +>; +def S_MIN_U32 : SOP2_32 <0x00000007, "S_MIN_U32", + [(set i32:$dst, (AMDGPUumin i32:$src0, i32:$src1))] +>; +def S_MAX_I32 : SOP2_32 <0x00000008, "S_MAX_I32", + [(set i32:$dst, (AMDGPUsmax i32:$src0, i32:$src1))] +>; +def S_MAX_U32 : SOP2_32 <0x00000009, "S_MAX_U32", + [(set i32:$dst, (AMDGPUumax i32:$src0, i32:$src1))] +>; + +def S_CSELECT_B32 : SOP2 < + 0x0000000a, (outs SReg_32:$dst), + (ins SReg_32:$src0, SReg_32:$src1, SCCReg:$scc), "S_CSELECT_B32", + [] +>; + +def S_CSELECT_B64 : SOP2_64 <0x0000000b, "S_CSELECT_B64", []>; + +def S_AND_B32 : SOP2_32 <0x0000000e, "S_AND_B32", + [(set i32:$dst, (and i32:$src0, i32:$src1))] +>; + +def S_AND_B64 : SOP2_64 <0x0000000f, "S_AND_B64", + [(set i64:$dst, (and i64:$src0, i64:$src1))] +>; + +def : Pat < + (i1 (and i1:$src0, i1:$src1)), + (S_AND_B64 $src0, $src1) +>; + +def S_OR_B32 : SOP2_32 <0x00000010, "S_OR_B32", + [(set i32:$dst, (or i32:$src0, i32:$src1))] +>; + +def S_OR_B64 : SOP2_64 <0x00000011, "S_OR_B64", + [(set i64:$dst, (or i64:$src0, i64:$src1))] +>; + +def : Pat < + (i1 (or i1:$src0, i1:$src1)), + (S_OR_B64 $src0, $src1) +>; + +def S_XOR_B32 : SOP2_32 <0x00000012, "S_XOR_B32", + [(set i32:$dst, (xor i32:$src0, i32:$src1))] +>; + +def S_XOR_B64 : SOP2_64 <0x00000013, "S_XOR_B64", + [(set i1:$dst, (xor i1:$src0, i1:$src1))] +>; +def S_ANDN2_B32 : SOP2_32 <0x00000014, "S_ANDN2_B32", []>; +def S_ANDN2_B64 : SOP2_64 <0x00000015, "S_ANDN2_B64", []>; +def S_ORN2_B32 : SOP2_32 <0x00000016, "S_ORN2_B32", []>; +def S_ORN2_B64 : SOP2_64 <0x00000017, "S_ORN2_B64", []>; +def S_NAND_B32 : SOP2_32 <0x00000018, "S_NAND_B32", []>; +def S_NAND_B64 : SOP2_64 <0x00000019, "S_NAND_B64", []>; +def S_NOR_B32 : SOP2_32 <0x0000001a, "S_NOR_B32", []>; +def S_NOR_B64 : SOP2_64 <0x0000001b, "S_NOR_B64", []>; +def S_XNOR_B32 : SOP2_32 <0x0000001c, "S_XNOR_B32", []>; +def S_XNOR_B64 : SOP2_64 <0x0000001d, "S_XNOR_B64", []>; + +// Use added complexity so these patterns are preferred to the VALU patterns. +let AddedComplexity = 1 in { + +def S_LSHL_B32 : SOP2_32 <0x0000001e, "S_LSHL_B32", + [(set i32:$dst, (shl i32:$src0, i32:$src1))] +>; +def S_LSHL_B64 : SOP2_SHIFT_64 <0x0000001f, "S_LSHL_B64", + [(set i64:$dst, (shl i64:$src0, i32:$src1))] +>; +def S_LSHR_B32 : SOP2_32 <0x00000020, "S_LSHR_B32", + [(set i32:$dst, (srl i32:$src0, i32:$src1))] +>; +def S_LSHR_B64 : SOP2_SHIFT_64 <0x00000021, "S_LSHR_B64", + [(set i64:$dst, (srl i64:$src0, i32:$src1))] +>; +def S_ASHR_I32 : SOP2_32 <0x00000022, "S_ASHR_I32", + [(set i32:$dst, (sra i32:$src0, i32:$src1))] +>; +def S_ASHR_I64 : SOP2_SHIFT_64 <0x00000023, "S_ASHR_I64", + [(set i64:$dst, (sra i64:$src0, i32:$src1))] +>; + +} // End AddedComplexity = 1 + +def S_BFM_B32 : SOP2_32 <0x00000024, "S_BFM_B32", []>; +def S_BFM_B64 : SOP2_64 <0x00000025, "S_BFM_B64", []>; +def S_MUL_I32 : SOP2_32 <0x00000026, "S_MUL_I32", []>; +def S_BFE_U32 : SOP2_32 <0x00000027, "S_BFE_U32", []>; +def S_BFE_I32 : SOP2_32 <0x00000028, "S_BFE_I32", []>; +def S_BFE_U64 : SOP2_64 <0x00000029, "S_BFE_U64", []>; +def S_BFE_I64 : SOP2_64 <0x0000002a, "S_BFE_I64", []>; +//def S_CBRANCH_G_FORK : SOP2_ <0x0000002b, "S_CBRANCH_G_FORK", []>; +def S_ABSDIFF_I32 : SOP2_32 <0x0000002c, "S_ABSDIFF_I32", []>; + +//===----------------------------------------------------------------------===// +// SOPC Instructions +//===----------------------------------------------------------------------===// + +def S_CMP_EQ_I32 : SOPC_32 <0x00000000, "S_CMP_EQ_I32">; +def S_CMP_LG_I32 : SOPC_32 <0x00000001, "S_CMP_LG_I32">; +def S_CMP_GT_I32 : SOPC_32 <0x00000002, "S_CMP_GT_I32">; +def S_CMP_GE_I32 : SOPC_32 <0x00000003, "S_CMP_GE_I32">; +def S_CMP_LT_I32 : SOPC_32 <0x00000004, "S_CMP_LT_I32">; +def S_CMP_LE_I32 : SOPC_32 <0x00000005, "S_CMP_LE_I32">; +def S_CMP_EQ_U32 : SOPC_32 <0x00000006, "S_CMP_EQ_U32">; +def S_CMP_LG_U32 : SOPC_32 <0x00000007, "S_CMP_LG_U32">; +def S_CMP_GT_U32 : SOPC_32 <0x00000008, "S_CMP_GT_U32">; +def S_CMP_GE_U32 : SOPC_32 <0x00000009, "S_CMP_GE_U32">; +def S_CMP_LT_U32 : SOPC_32 <0x0000000a, "S_CMP_LT_U32">; +def S_CMP_LE_U32 : SOPC_32 <0x0000000b, "S_CMP_LE_U32">; +////def S_BITCMP0_B32 : SOPC_BITCMP0 <0x0000000c, "S_BITCMP0_B32", []>; +////def S_BITCMP1_B32 : SOPC_BITCMP1 <0x0000000d, "S_BITCMP1_B32", []>; +////def S_BITCMP0_B64 : SOPC_BITCMP0 <0x0000000e, "S_BITCMP0_B64", []>; +////def S_BITCMP1_B64 : SOPC_BITCMP1 <0x0000000f, "S_BITCMP1_B64", []>; +//def S_SETVSKIP : SOPC_ <0x00000010, "S_SETVSKIP", []>; + +//===----------------------------------------------------------------------===// +// SOPK Instructions +//===----------------------------------------------------------------------===// + def S_MOVK_I32 : SOPK_32 <0x00000000, "S_MOVK_I32", []>; def S_CMOVK_I32 : SOPK_32 <0x00000002, "S_CMOVK_I32", []>; @@ -155,6 +353,106 @@ def S_GETREG_REGRD_B32 : SOPK_32 <0x00000014, "S_GETREG_REGRD_B32", []>; //def S_SETREG_IMM32_B32 : SOPK_32 <0x00000015, "S_SETREG_IMM32_B32", []>; //def EXP : EXP_ <0x00000000, "EXP", []>; +//===----------------------------------------------------------------------===// +// SOPP Instructions +//===----------------------------------------------------------------------===// + +//def S_NOP : SOPP_ <0x00000000, "S_NOP", []>; + +let isTerminator = 1 in { + +def S_ENDPGM : SOPP <0x00000001, (ins), "S_ENDPGM", + [(IL_retflag)]> { + let SIMM16 = 0; + let isBarrier = 1; + let hasCtrlDep = 1; +} + +let isBranch = 1 in { +def S_BRANCH : SOPP < + 0x00000002, (ins brtarget:$target), "S_BRANCH $target", + [(br bb:$target)]> { + let isBarrier = 1; +} + +let DisableEncoding = "$scc" in { +def S_CBRANCH_SCC0 : SOPP < + 0x00000004, (ins brtarget:$target, SCCReg:$scc), + "S_CBRANCH_SCC0 $target", [] +>; +def S_CBRANCH_SCC1 : SOPP < + 0x00000005, (ins brtarget:$target, SCCReg:$scc), + "S_CBRANCH_SCC1 $target", + [] +>; +} // End DisableEncoding = "$scc" + +def S_CBRANCH_VCCZ : SOPP < + 0x00000006, (ins brtarget:$target, VCCReg:$vcc), + "S_CBRANCH_VCCZ $target", + [] +>; +def S_CBRANCH_VCCNZ : SOPP < + 0x00000007, (ins brtarget:$target, VCCReg:$vcc), + "S_CBRANCH_VCCNZ $target", + [] +>; + +let DisableEncoding = "$exec" in { +def S_CBRANCH_EXECZ : SOPP < + 0x00000008, (ins brtarget:$target, EXECReg:$exec), + "S_CBRANCH_EXECZ $target", + [] +>; +def S_CBRANCH_EXECNZ : SOPP < + 0x00000009, (ins brtarget:$target, EXECReg:$exec), + "S_CBRANCH_EXECNZ $target", + [] +>; +} // End DisableEncoding = "$exec" + + +} // End isBranch = 1 +} // End isTerminator = 1 + +let hasSideEffects = 1 in { +def S_BARRIER : SOPP <0x0000000a, (ins), "S_BARRIER", + [(int_AMDGPU_barrier_local)] +> { + let SIMM16 = 0; + let isBarrier = 1; + let hasCtrlDep = 1; + let mayLoad = 1; + let mayStore = 1; +} + +def S_WAITCNT : SOPP <0x0000000c, (ins WAIT_FLAG:$simm16), "S_WAITCNT $simm16", + [] +>; +//def S_SETHALT : SOPP_ <0x0000000d, "S_SETHALT", []>; +//def S_SLEEP : SOPP_ <0x0000000e, "S_SLEEP", []>; +//def S_SETPRIO : SOPP_ <0x0000000f, "S_SETPRIO", []>; + +let Uses = [EXEC] in { + def S_SENDMSG : SOPP <0x00000010, (ins SendMsgImm:$simm16, M0Reg:$m0), "S_SENDMSG $simm16", + [(int_SI_sendmsg imm:$simm16, M0Reg:$m0)] + > { + let DisableEncoding = "$m0"; + } +} // End Uses = [EXEC] + +//def S_SENDMSGHALT : SOPP_ <0x00000011, "S_SENDMSGHALT", []>; +//def S_TRAP : SOPP_ <0x00000012, "S_TRAP", []>; +//def S_ICACHE_INV : SOPP_ <0x00000013, "S_ICACHE_INV", []>; +//def S_INCPERFLEVEL : SOPP_ <0x00000014, "S_INCPERFLEVEL", []>; +//def S_DECPERFLEVEL : SOPP_ <0x00000015, "S_DECPERFLEVEL", []>; +//def S_TTRACEDATA : SOPP_ <0x00000016, "S_TTRACEDATA", []>; +} // End hasSideEffects + +//===----------------------------------------------------------------------===// +// VOPC Instructions +//===----------------------------------------------------------------------===// + let isCompare = 1 in { defm V_CMP_F_F32 : VOPC_32 <0x00000000, "V_CMP_F_F32">; @@ -411,6 +709,10 @@ defm V_CMPX_CLASS_F64 : VOPC_64 <0x000000b8, "V_CMPX_CLASS_F64">; } // End isCompare = 1 +//===----------------------------------------------------------------------===// +// DS Instructions +//===----------------------------------------------------------------------===// + def DS_ADD_U32_RTN : DS_1A1D_RET <0x20, "DS_ADD_U32_RTN", VReg_32>; def DS_SUB_U32_RTN : DS_1A1D_RET <0x21, "DS_SUB_U32_RTN", VReg_32>; def DS_WRITE_B32 : DS_Store_Helper <0x0000000d, "DS_WRITE_B32", VReg_32>; @@ -435,6 +737,9 @@ def DS_READ2_B64 : DS_Load2_Helper <0x00000075, "DS_READ2_B64", VReg_128>; // TODO: DS_READ2ST64_B32, DS_READ2ST64_B64, // DS_WRITE2ST64_B32, DS_WRITE2ST64_B64 +//===----------------------------------------------------------------------===// +// MUBUF Instructions +//===----------------------------------------------------------------------===// //def BUFFER_LOAD_FORMAT_X : MUBUF_ <0x00000000, "BUFFER_LOAD_FORMAT_X", []>; //def BUFFER_LOAD_FORMAT_XY : MUBUF_ <0x00000001, "BUFFER_LOAD_FORMAT_XY", []>; @@ -507,6 +812,11 @@ def BUFFER_STORE_DWORDX4 : MUBUF_Store_Helper < //def BUFFER_ATOMIC_FMAX_X2 : MUBUF_X2 <0x00000060, "BUFFER_ATOMIC_FMAX_X2", []>; //def BUFFER_WBINVL1_SC : MUBUF_WBINVL1 <0x00000070, "BUFFER_WBINVL1_SC", []>; //def BUFFER_WBINVL1 : MUBUF_WBINVL1 <0x00000071, "BUFFER_WBINVL1", []>; + +//===----------------------------------------------------------------------===// +// MTBUF Instructions +//===----------------------------------------------------------------------===// + //def TBUFFER_LOAD_FORMAT_X : MTBUF_ <0x00000000, "TBUFFER_LOAD_FORMAT_X", []>; //def TBUFFER_LOAD_FORMAT_XY : MTBUF_ <0x00000001, "TBUFFER_LOAD_FORMAT_XY", []>; //def TBUFFER_LOAD_FORMAT_XYZ : MTBUF_ <0x00000002, "TBUFFER_LOAD_FORMAT_XYZ", []>; @@ -516,41 +826,10 @@ def TBUFFER_STORE_FORMAT_XY : MTBUF_Store_Helper <0x00000005, "TBUFFER_STORE_FOR def TBUFFER_STORE_FORMAT_XYZ : MTBUF_Store_Helper <0x00000006, "TBUFFER_STORE_FORMAT_XYZ", VReg_128>; def TBUFFER_STORE_FORMAT_XYZW : MTBUF_Store_Helper <0x00000007, "TBUFFER_STORE_FORMAT_XYZW", VReg_128>; -let mayLoad = 1 in { +//===----------------------------------------------------------------------===// +// MIMG Instructions +//===----------------------------------------------------------------------===// -// We are using the SGPR_32 and not the SReg_32 register class for 32-bit -// SMRD instructions, because the SGPR_32 register class does not include M0 -// and writing to M0 from an SMRD instruction will hang the GPU. -defm S_LOAD_DWORD : SMRD_Helper <0x00, "S_LOAD_DWORD", SReg_64, SGPR_32>; -defm S_LOAD_DWORDX2 : SMRD_Helper <0x01, "S_LOAD_DWORDX2", SReg_64, SReg_64>; -defm S_LOAD_DWORDX4 : SMRD_Helper <0x02, "S_LOAD_DWORDX4", SReg_64, SReg_128>; -defm S_LOAD_DWORDX8 : SMRD_Helper <0x03, "S_LOAD_DWORDX8", SReg_64, SReg_256>; -defm S_LOAD_DWORDX16 : SMRD_Helper <0x04, "S_LOAD_DWORDX16", SReg_64, SReg_512>; - -defm S_BUFFER_LOAD_DWORD : SMRD_Helper < - 0x08, "S_BUFFER_LOAD_DWORD", SReg_128, SGPR_32 ->; - -defm S_BUFFER_LOAD_DWORDX2 : SMRD_Helper < - 0x09, "S_BUFFER_LOAD_DWORDX2", SReg_128, SReg_64 ->; - -defm S_BUFFER_LOAD_DWORDX4 : SMRD_Helper < - 0x0a, "S_BUFFER_LOAD_DWORDX4", SReg_128, SReg_128 ->; - -defm S_BUFFER_LOAD_DWORDX8 : SMRD_Helper < - 0x0b, "S_BUFFER_LOAD_DWORDX8", SReg_128, SReg_256 ->; - -defm S_BUFFER_LOAD_DWORDX16 : SMRD_Helper < - 0x0c, "S_BUFFER_LOAD_DWORDX16", SReg_128, SReg_512 ->; - -} // mayLoad = 1 - -//def S_MEMTIME : SMRD_ <0x0000001e, "S_MEMTIME", []>; -//def S_DCACHE_INV : SMRD_ <0x0000001f, "S_DCACHE_INV", []>; defm IMAGE_LOAD : MIMG_NoSampler <0x00000000, "IMAGE_LOAD">; defm IMAGE_LOAD_MIP : MIMG_NoSampler <0x00000001, "IMAGE_LOAD_MIP">; //def IMAGE_LOAD_PCK : MIMG_NoPattern_ <"IMAGE_LOAD_PCK", 0x00000002>; @@ -646,8 +925,12 @@ defm IMAGE_SAMPLE_C_B : MIMG_Sampler <0x0000002d, "IMAGE_SAMPLE_C_B">; //def IMAGE_SAMPLE_C_CD_CL_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_CD_CL_O", 0x0000006f>; //def IMAGE_RSRC256 : MIMG_NoPattern_RSRC256 <"IMAGE_RSRC256", 0x0000007e>; //def IMAGE_SAMPLER : MIMG_NoPattern_ <"IMAGE_SAMPLER", 0x0000007f>; -//def V_NOP : VOP1_ <0x00000000, "V_NOP", []>; +//===----------------------------------------------------------------------===// +// VOP1 Instructions +//===----------------------------------------------------------------------===// + +//def V_NOP : VOP1_ <0x00000000, "V_NOP", []>; let neverHasSideEffects = 1, isMoveImm = 1 in { defm V_MOV_B32 : VOP1_32 <0x00000001, "V_MOV_B32", []>; @@ -764,6 +1047,11 @@ defm V_MOVRELD_B32 : VOP1_32 <0x00000042, "V_MOVRELD_B32", []>; defm V_MOVRELS_B32 : VOP1_32 <0x00000043, "V_MOVRELS_B32", []>; defm V_MOVRELSD_B32 : VOP1_32 <0x00000044, "V_MOVRELSD_B32", []>; + +//===----------------------------------------------------------------------===// +// VINTRP Instructions +//===----------------------------------------------------------------------===// + def V_INTERP_P1_F32 : VINTRP < 0x00000000, (outs VReg_32:$dst), @@ -794,97 +1082,9 @@ def V_INTERP_MOV_F32 : VINTRP < let DisableEncoding = "$m0"; } -//def S_NOP : SOPP_ <0x00000000, "S_NOP", []>; - -let isTerminator = 1 in { - -def S_ENDPGM : SOPP <0x00000001, (ins), "S_ENDPGM", - [(IL_retflag)]> { - let SIMM16 = 0; - let isBarrier = 1; - let hasCtrlDep = 1; -} - -let isBranch = 1 in { -def S_BRANCH : SOPP < - 0x00000002, (ins brtarget:$target), "S_BRANCH $target", - [(br bb:$target)]> { - let isBarrier = 1; -} - -let DisableEncoding = "$scc" in { -def S_CBRANCH_SCC0 : SOPP < - 0x00000004, (ins brtarget:$target, SCCReg:$scc), - "S_CBRANCH_SCC0 $target", [] ->; -def S_CBRANCH_SCC1 : SOPP < - 0x00000005, (ins brtarget:$target, SCCReg:$scc), - "S_CBRANCH_SCC1 $target", - [] ->; -} // End DisableEncoding = "$scc" - -def S_CBRANCH_VCCZ : SOPP < - 0x00000006, (ins brtarget:$target, VCCReg:$vcc), - "S_CBRANCH_VCCZ $target", - [] ->; -def S_CBRANCH_VCCNZ : SOPP < - 0x00000007, (ins brtarget:$target, VCCReg:$vcc), - "S_CBRANCH_VCCNZ $target", - [] ->; - -let DisableEncoding = "$exec" in { -def S_CBRANCH_EXECZ : SOPP < - 0x00000008, (ins brtarget:$target, EXECReg:$exec), - "S_CBRANCH_EXECZ $target", - [] ->; -def S_CBRANCH_EXECNZ : SOPP < - 0x00000009, (ins brtarget:$target, EXECReg:$exec), - "S_CBRANCH_EXECNZ $target", - [] ->; -} // End DisableEncoding = "$exec" - - -} // End isBranch = 1 -} // End isTerminator = 1 - -let hasSideEffects = 1 in { -def S_BARRIER : SOPP <0x0000000a, (ins), "S_BARRIER", - [(int_AMDGPU_barrier_local)] -> { - let SIMM16 = 0; - let isBarrier = 1; - let hasCtrlDep = 1; - let mayLoad = 1; - let mayStore = 1; -} - -def S_WAITCNT : SOPP <0x0000000c, (ins WAIT_FLAG:$simm16), "S_WAITCNT $simm16", - [] ->; -//def S_SETHALT : SOPP_ <0x0000000d, "S_SETHALT", []>; -//def S_SLEEP : SOPP_ <0x0000000e, "S_SLEEP", []>; -//def S_SETPRIO : SOPP_ <0x0000000f, "S_SETPRIO", []>; - -let Uses = [EXEC] in { - def S_SENDMSG : SOPP <0x00000010, (ins SendMsgImm:$simm16, M0Reg:$m0), "S_SENDMSG $simm16", - [(int_SI_sendmsg imm:$simm16, M0Reg:$m0)] - > { - let DisableEncoding = "$m0"; - } -} // End Uses = [EXEC] - -//def S_SENDMSGHALT : SOPP_ <0x00000011, "S_SENDMSGHALT", []>; -//def S_TRAP : SOPP_ <0x00000012, "S_TRAP", []>; -//def S_ICACHE_INV : SOPP_ <0x00000013, "S_ICACHE_INV", []>; -//def S_INCPERFLEVEL : SOPP_ <0x00000014, "S_INCPERFLEVEL", []>; -//def S_DECPERFLEVEL : SOPP_ <0x00000015, "S_DECPERFLEVEL", []>; -//def S_TTRACEDATA : SOPP_ <0x00000016, "S_TTRACEDATA", []>; -} // End hasSideEffects +//===----------------------------------------------------------------------===// +// VOP2 Instructions +//===----------------------------------------------------------------------===// def V_CNDMASK_B32_e32 : VOP2 <0x00000000, (outs VReg_32:$dst), (ins VSrc_32:$src0, VReg_32:$src1, VCCReg:$vcc), @@ -1031,23 +1231,10 @@ defm V_CVT_PKRTZ_F16_F32 : VOP2_32 <0x0000002f, "V_CVT_PKRTZ_F16_F32", >; ////def V_CVT_PK_U16_U32 : VOP2_U16 <0x00000030, "V_CVT_PK_U16_U32", []>; ////def V_CVT_PK_I16_I32 : VOP2_I16 <0x00000031, "V_CVT_PK_I16_I32", []>; -def S_CMP_EQ_I32 : SOPC_32 <0x00000000, "S_CMP_EQ_I32">; -def S_CMP_LG_I32 : SOPC_32 <0x00000001, "S_CMP_LG_I32">; -def S_CMP_GT_I32 : SOPC_32 <0x00000002, "S_CMP_GT_I32">; -def S_CMP_GE_I32 : SOPC_32 <0x00000003, "S_CMP_GE_I32">; -def S_CMP_LT_I32 : SOPC_32 <0x00000004, "S_CMP_LT_I32">; -def S_CMP_LE_I32 : SOPC_32 <0x00000005, "S_CMP_LE_I32">; -def S_CMP_EQ_U32 : SOPC_32 <0x00000006, "S_CMP_EQ_U32">; -def S_CMP_LG_U32 : SOPC_32 <0x00000007, "S_CMP_LG_U32">; -def S_CMP_GT_U32 : SOPC_32 <0x00000008, "S_CMP_GT_U32">; -def S_CMP_GE_U32 : SOPC_32 <0x00000009, "S_CMP_GE_U32">; -def S_CMP_LT_U32 : SOPC_32 <0x0000000a, "S_CMP_LT_U32">; -def S_CMP_LE_U32 : SOPC_32 <0x0000000b, "S_CMP_LE_U32">; -////def S_BITCMP0_B32 : SOPC_BITCMP0 <0x0000000c, "S_BITCMP0_B32", []>; -////def S_BITCMP1_B32 : SOPC_BITCMP1 <0x0000000d, "S_BITCMP1_B32", []>; -////def S_BITCMP0_B64 : SOPC_BITCMP0 <0x0000000e, "S_BITCMP0_B64", []>; -////def S_BITCMP1_B64 : SOPC_BITCMP1 <0x0000000f, "S_BITCMP1_B64", []>; -//def S_SETVSKIP : SOPC_ <0x00000010, "S_SETVSKIP", []>; + +//===----------------------------------------------------------------------===// +// VOP3 Instructions +//===----------------------------------------------------------------------===// let neverHasSideEffects = 1 in { @@ -1169,128 +1356,9 @@ def V_DIV_FMAS_F64 : VOP3_64 <0x00000170, "V_DIV_FMAS_F64", []>; //def V_MQSAD_U8 : VOP3_U8 <0x00000173, "V_MQSAD_U8", []>; def V_TRIG_PREOP_F64 : VOP3_64 <0x00000174, "V_TRIG_PREOP_F64", []>; -let Defs = [SCC] in { // Carry out goes to SCC -let isCommutable = 1 in { -def S_ADD_U32 : SOP2_32 <0x00000000, "S_ADD_U32", []>; -def S_ADD_I32 : SOP2_32 <0x00000002, "S_ADD_I32", - [(set i32:$dst, (add SSrc_32:$src0, SSrc_32:$src1))] ->; -} // End isCommutable = 1 - -def S_SUB_U32 : SOP2_32 <0x00000001, "S_SUB_U32", []>; -def S_SUB_I32 : SOP2_32 <0x00000003, "S_SUB_I32", - [(set i32:$dst, (sub SSrc_32:$src0, SSrc_32:$src1))] ->; - -let Uses = [SCC] in { // Carry in comes from SCC -let isCommutable = 1 in { -def S_ADDC_U32 : SOP2_32 <0x00000004, "S_ADDC_U32", - [(set i32:$dst, (adde (i32 SSrc_32:$src0), (i32 SSrc_32:$src1)))]>; -} // End isCommutable = 1 - -def S_SUBB_U32 : SOP2_32 <0x00000005, "S_SUBB_U32", - [(set i32:$dst, (sube (i32 SSrc_32:$src0), (i32 SSrc_32:$src1)))]>; -} // End Uses = [SCC] -} // End Defs = [SCC] - -def S_MIN_I32 : SOP2_32 <0x00000006, "S_MIN_I32", - [(set i32:$dst, (AMDGPUsmin i32:$src0, i32:$src1))] ->; -def S_MIN_U32 : SOP2_32 <0x00000007, "S_MIN_U32", - [(set i32:$dst, (AMDGPUumin i32:$src0, i32:$src1))] ->; -def S_MAX_I32 : SOP2_32 <0x00000008, "S_MAX_I32", - [(set i32:$dst, (AMDGPUsmax i32:$src0, i32:$src1))] ->; -def S_MAX_U32 : SOP2_32 <0x00000009, "S_MAX_U32", - [(set i32:$dst, (AMDGPUumax i32:$src0, i32:$src1))] ->; - -def S_CSELECT_B32 : SOP2 < - 0x0000000a, (outs SReg_32:$dst), - (ins SReg_32:$src0, SReg_32:$src1, SCCReg:$scc), "S_CSELECT_B32", - [] ->; - -def S_CSELECT_B64 : SOP2_64 <0x0000000b, "S_CSELECT_B64", []>; - -def S_AND_B32 : SOP2_32 <0x0000000e, "S_AND_B32", - [(set i32:$dst, (and i32:$src0, i32:$src1))] ->; - -def S_AND_B64 : SOP2_64 <0x0000000f, "S_AND_B64", - [(set i64:$dst, (and i64:$src0, i64:$src1))] ->; - -def : Pat < - (i1 (and i1:$src0, i1:$src1)), - (S_AND_B64 $src0, $src1) ->; - -def S_OR_B32 : SOP2_32 <0x00000010, "S_OR_B32", - [(set i32:$dst, (or i32:$src0, i32:$src1))] ->; - -def S_OR_B64 : SOP2_64 <0x00000011, "S_OR_B64", - [(set i64:$dst, (or i64:$src0, i64:$src1))] ->; - -def : Pat < - (i1 (or i1:$src0, i1:$src1)), - (S_OR_B64 $src0, $src1) ->; - -def S_XOR_B32 : SOP2_32 <0x00000012, "S_XOR_B32", - [(set i32:$dst, (xor i32:$src0, i32:$src1))] ->; - -def S_XOR_B64 : SOP2_64 <0x00000013, "S_XOR_B64", - [(set i1:$dst, (xor i1:$src0, i1:$src1))] ->; -def S_ANDN2_B32 : SOP2_32 <0x00000014, "S_ANDN2_B32", []>; -def S_ANDN2_B64 : SOP2_64 <0x00000015, "S_ANDN2_B64", []>; -def S_ORN2_B32 : SOP2_32 <0x00000016, "S_ORN2_B32", []>; -def S_ORN2_B64 : SOP2_64 <0x00000017, "S_ORN2_B64", []>; -def S_NAND_B32 : SOP2_32 <0x00000018, "S_NAND_B32", []>; -def S_NAND_B64 : SOP2_64 <0x00000019, "S_NAND_B64", []>; -def S_NOR_B32 : SOP2_32 <0x0000001a, "S_NOR_B32", []>; -def S_NOR_B64 : SOP2_64 <0x0000001b, "S_NOR_B64", []>; -def S_XNOR_B32 : SOP2_32 <0x0000001c, "S_XNOR_B32", []>; -def S_XNOR_B64 : SOP2_64 <0x0000001d, "S_XNOR_B64", []>; - -// Use added complexity so these patterns are preferred to the VALU patterns. -let AddedComplexity = 1 in { - -def S_LSHL_B32 : SOP2_32 <0x0000001e, "S_LSHL_B32", - [(set i32:$dst, (shl i32:$src0, i32:$src1))] ->; -def S_LSHL_B64 : SOP2_SHIFT_64 <0x0000001f, "S_LSHL_B64", - [(set i64:$dst, (shl i64:$src0, i32:$src1))] ->; -def S_LSHR_B32 : SOP2_32 <0x00000020, "S_LSHR_B32", - [(set i32:$dst, (srl i32:$src0, i32:$src1))] ->; -def S_LSHR_B64 : SOP2_SHIFT_64 <0x00000021, "S_LSHR_B64", - [(set i64:$dst, (srl i64:$src0, i32:$src1))] ->; -def S_ASHR_I32 : SOP2_32 <0x00000022, "S_ASHR_I32", - [(set i32:$dst, (sra i32:$src0, i32:$src1))] ->; -def S_ASHR_I64 : SOP2_SHIFT_64 <0x00000023, "S_ASHR_I64", - [(set i64:$dst, (sra i64:$src0, i32:$src1))] ->; - -} // End AddedComplexity = 1 - -def S_BFM_B32 : SOP2_32 <0x00000024, "S_BFM_B32", []>; -def S_BFM_B64 : SOP2_64 <0x00000025, "S_BFM_B64", []>; -def S_MUL_I32 : SOP2_32 <0x00000026, "S_MUL_I32", []>; -def S_BFE_U32 : SOP2_32 <0x00000027, "S_BFE_U32", []>; -def S_BFE_I32 : SOP2_32 <0x00000028, "S_BFE_I32", []>; -def S_BFE_U64 : SOP2_64 <0x00000029, "S_BFE_U64", []>; -def S_BFE_I64 : SOP2_64 <0x0000002a, "S_BFE_I64", []>; -//def S_CBRANCH_G_FORK : SOP2_ <0x0000002b, "S_CBRANCH_G_FORK", []>; -def S_ABSDIFF_I32 : SOP2_32 <0x0000002c, "S_ABSDIFF_I32", []>; +//===----------------------------------------------------------------------===// +// Pseudo Instructions +//===----------------------------------------------------------------------===// let isCodeGenOnly = 1, isPseudo = 1 in { @@ -1478,6 +1546,53 @@ def : Pat < (V_SUB_F64 $src0, $src1) >; +//===----------------------------------------------------------------------===// +// SMRD Patterns +//===----------------------------------------------------------------------===// + +multiclass SMRD_Pattern { + + // 1. Offset as 8bit DWORD immediate + def : Pat < + (constant_load (add i64:$sbase, (i64 IMM8bitDWORD:$offset))), + (vt (Instr_IMM $sbase, (as_dword_i32imm $offset))) + >; + + // 2. Offset loaded in an 32bit SGPR + def : Pat < + (constant_load (SIadd64bit32bit i64:$sbase, imm:$offset)), + (vt (Instr_SGPR $sbase, (S_MOV_B32 imm:$offset))) + >; + + // 3. No offset at all + def : Pat < + (constant_load i64:$sbase), + (vt (Instr_IMM $sbase, 0)) + >; +} + +defm : SMRD_Pattern ; +defm : SMRD_Pattern ; +defm : SMRD_Pattern ; +defm : SMRD_Pattern ; +defm : SMRD_Pattern ; +defm : SMRD_Pattern ; +defm : SMRD_Pattern ; +defm : SMRD_Pattern ; + +// 1. Offset as 8bit DWORD immediate +def : Pat < + (SIload_constant v4i32:$sbase, IMM8bitDWORD:$offset), + (S_BUFFER_LOAD_DWORD_IMM $sbase, (as_dword_i32imm $offset)) +>; + +// 2. Offset loaded in an 32bit SGPR +def : Pat < + (SIload_constant v4i32:$sbase, imm:$offset), + (S_BUFFER_LOAD_DWORD_SGPR $sbase, (S_MOV_B32 imm:$offset)) +>; + + /********** ======================= **********/ /********** Image sampling patterns **********/ /********** ======================= **********/ @@ -1866,19 +1981,7 @@ class Ext32Pat : Pat < def : Ext32Pat ; def : Ext32Pat ; -// 1. Offset as 8bit DWORD immediate -def : Pat < - (SIload_constant v4i32:$sbase, IMM8bitDWORD:$offset), - (S_BUFFER_LOAD_DWORD_IMM $sbase, (as_dword_i32imm $offset)) ->; - -// 2. Offset loaded in an 32bit SGPR -def : Pat < - (SIload_constant v4i32:$sbase, imm:$offset), - (S_BUFFER_LOAD_DWORD_SGPR $sbase, (S_MOV_B32 imm:$offset)) ->; - -// 3. Offset in an 32Bit VGPR +// Offset in an 32Bit VGPR def : Pat < (SIload_constant v4i32:$sbase, i32:$voff), (BUFFER_LOAD_DWORD_OFFEN $sbase, $voff, 0, 0, 0, 0) @@ -1953,40 +2056,6 @@ def : Pat <(atomic_load_add_local i32:$ptr, i32:$val), def : Pat <(atomic_load_sub_local i32:$ptr, i32:$val), (DS_SUB_U32_RTN 0, $ptr, $val, 0)>; -/********** ================== **********/ -/********** SMRD Patterns **********/ -/********** ================== **********/ - -multiclass SMRD_Pattern { - - // 1. Offset as 8bit DWORD immediate - def : Pat < - (constant_load (add i64:$sbase, (i64 IMM8bitDWORD:$offset))), - (vt (Instr_IMM $sbase, (as_dword_i32imm $offset))) - >; - - // 2. Offset loaded in an 32bit SGPR - def : Pat < - (constant_load (SIadd64bit32bit i64:$sbase, imm:$offset)), - (vt (Instr_SGPR $sbase, (S_MOV_B32 imm:$offset))) - >; - - // 3. No offset at all - def : Pat < - (constant_load i64:$sbase), - (vt (Instr_IMM $sbase, 0)) - >; -} - -defm : SMRD_Pattern ; -defm : SMRD_Pattern ; -defm : SMRD_Pattern ; -defm : SMRD_Pattern ; -defm : SMRD_Pattern ; -defm : SMRD_Pattern ; -defm : SMRD_Pattern ; -defm : SMRD_Pattern ; - //===----------------------------------------------------------------------===// // MUBUF Patterns //===----------------------------------------------------------------------===//