f9adee2f6b
Some checks failed
Bazel Checks / Buildifier (push) Has been cancelled
Bazel Checks / Bazel Build/Test (push) Has been cancelled
Build CI Tooling Containers / Build Container abi-tests (push) Has been cancelled
Build CI Tooling Containers / Build Container format (push) Has been cancelled
Build CI Tooling Containers / Build Container lint (push) Has been cancelled
Build Windows CI Container / build-ci-container-windows (push) Has been cancelled
Build CI Container / Build Container X64 (push) Has been cancelled
Build CI Container / Build Container ARM64 (push) Has been cancelled
Build CI Container / Build Container agent X64 (push) Has been cancelled
Build CI Container / Build Container agent ARM64 (push) Has been cancelled
Build libc Container / Build libc container (ubuntu-24.04) (push) Has been cancelled
Build libc Container / Build libc container (ubuntu-24.04-arm) (push) Has been cancelled
Build Metrics Container / build-metrics-container (push) Has been cancelled
Check CI Scripts / Check Python Tests (push) Has been cancelled
Test documentation build / Test documentation build (push) Has been cancelled
Libclang Python Binding Tests / Build and run Python unit tests (3.13) (push) Has been cancelled
Libclang Python Binding Tests / Build and run Python unit tests (3.8) (push) Has been cancelled
Build Docker images for libc++ CI / build-and-push (push) Has been cancelled
Test Unprivileged Download Artifact Action / Upload Test Artifact (push) Has been cancelled
Zizmor GitHub Actions Analysis / Run zizmor (push) Has been cancelled
Build CI Tooling Containers / push-ci-container (push) Has been cancelled
Build Windows CI Container / push-ci-container (push) Has been cancelled
Build CI Container / push-ci-container (push) Has been cancelled
Build libc Container / push-libc-container (push) Has been cancelled
Build Metrics Container / push-metrics-container (push) Has been cancelled
Test Unprivileged Download Artifact Action / Test Unprivileged Download Artifact (push) Has been cancelled
Commit Access Review / commit-access-review (push) Has been cancelled
The ASYNC_CNT is used to track the progress of asynchronous copies between global and LDS memories. By including it in asyncmark, the compiler can now assist the programmer in generating waits for ASYNC_CNT. Assisted-By: Claude Sonnet 4.5 This is part of a stack: - #185813 - #185810 Fixes: LCOMPILER-332
3534 lines
138 KiB
TableGen
3534 lines
138 KiB
TableGen
//===-- SOPInstructions.td - SOP Instruction Definitions ------------------===//
|
|
//
|
|
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
|
|
// See https://llvm.org/LICENSE.txt for license information.
|
|
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
|
|
//
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
def GPRIdxMode : CustomOperand;
|
|
|
|
class SOP_Pseudo<string opName, dag outs, dag ins, string asmOps,
|
|
list<dag> pattern=[]> :
|
|
InstSI<outs, ins, "", pattern>,
|
|
SIMCInstr<opName, SIEncodingFamily.NONE> {
|
|
|
|
let isPseudo = 1;
|
|
let isCodeGenOnly = 1;
|
|
let Size = 4;
|
|
|
|
string Mnemonic = opName;
|
|
string AsmOperands = asmOps;
|
|
|
|
bits<1> has_sdst = 0;
|
|
}
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOP1 Instructions
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
class SOP1_Pseudo <string opName, dag outs, dag ins,
|
|
string asmOps, list<dag> pattern=[]> :
|
|
SOP_Pseudo<opName, outs, ins, " " # asmOps, pattern> {
|
|
|
|
let mayLoad = 0;
|
|
let mayStore = 0;
|
|
let hasSideEffects = 0;
|
|
let SALU = 1;
|
|
let SOP1 = 1;
|
|
let SchedRW = [WriteSALU];
|
|
let UseNamedOperandTable = 1;
|
|
|
|
bits<1> has_src0 = 1;
|
|
let has_sdst = 1;
|
|
}
|
|
|
|
class SOP1_Real<bits<8> op, SOP1_Pseudo ps, string real_name = ps.Mnemonic> :
|
|
InstSI <ps.OutOperandList, ps.InOperandList,
|
|
real_name # ps.AsmOperands>,
|
|
Enc32 {
|
|
|
|
let SALU = 1;
|
|
let SOP1 = 1;
|
|
let isPseudo = 0;
|
|
let isCodeGenOnly = 0;
|
|
let Size = 4;
|
|
|
|
// copy relevant pseudo op flags
|
|
let SubtargetPredicate = ps.SubtargetPredicate;
|
|
let AsmMatchConverter = ps.AsmMatchConverter;
|
|
let SchedRW = ps.SchedRW;
|
|
let mayLoad = ps.mayLoad;
|
|
let mayStore = ps.mayStore;
|
|
let isTerminator = ps.isTerminator;
|
|
let isReturn = ps.isReturn;
|
|
let isCall = ps.isCall;
|
|
let isBranch = ps.isBranch;
|
|
let isBarrier = ps.isBarrier;
|
|
let Uses = ps.Uses;
|
|
let Defs = ps.Defs;
|
|
let isConvergent = ps.isConvergent;
|
|
|
|
// encoding
|
|
bits<7> sdst;
|
|
bits<8> src0;
|
|
|
|
let Inst{7-0} = !if(ps.has_src0, src0, ?);
|
|
let Inst{15-8} = op;
|
|
let Inst{22-16} = !if(ps.has_sdst, sdst, ?);
|
|
let Inst{31-23} = 0x17d; //encoding;
|
|
}
|
|
|
|
class SOP1_32 <string opName, list<dag> pattern=[], bit tied_in = 0> : SOP1_Pseudo <
|
|
opName, (outs SReg_32:$sdst),
|
|
!if(tied_in, (ins SSrc_b32:$src0, SReg_32:$sdst_in),
|
|
(ins SSrc_b32:$src0)),
|
|
"$sdst, $src0", pattern> {
|
|
let Constraints = !if(tied_in, "$sdst = $sdst_in", "");
|
|
}
|
|
|
|
// Only register input allowed.
|
|
class SOP1_32R <string opName, list<dag> pattern=[]> : SOP1_Pseudo <
|
|
opName, (outs SReg_32:$sdst), (ins SReg_32:$src0),
|
|
"$sdst, $src0", pattern>;
|
|
|
|
// 32-bit input, no output.
|
|
class SOP1_0_32 <string opName, list<dag> pattern = []> : SOP1_Pseudo <
|
|
opName, (outs), (ins SSrc_b32:$src0),
|
|
"$src0", pattern> {
|
|
let has_sdst = 0;
|
|
}
|
|
|
|
// Special case for movreld where sdst is treated as a use operand.
|
|
class SOP1_32_movreld <string opName, list<dag> pattern=[]> : SOP1_Pseudo <
|
|
opName, (outs), (ins SReg_32:$sdst, SSrc_b32:$src0),
|
|
"$sdst, $src0", pattern>;
|
|
|
|
// Special case for movreld where sdst is treated as a use operand.
|
|
class SOP1_64_movreld <string opName, list<dag> pattern=[]> : SOP1_Pseudo <
|
|
opName, (outs), (ins SReg_64:$sdst, SSrc_b64:$src0),
|
|
"$sdst, $src0", pattern
|
|
>;
|
|
|
|
class SOP1_0_32R <string opName, list<dag> pattern = []> : SOP1_Pseudo <
|
|
opName, (outs), (ins SReg_32:$src0),
|
|
"$src0", pattern> {
|
|
let has_sdst = 0;
|
|
}
|
|
|
|
class SOP1_64 <string opName, list<dag> pattern=[]> : SOP1_Pseudo <
|
|
opName, (outs SReg_64:$sdst), (ins SSrc_b64:$src0),
|
|
"$sdst, $src0", pattern
|
|
>;
|
|
|
|
// Only register input allowed.
|
|
class SOP1_64R <string opName, list<dag> pattern=[]> : SOP1_Pseudo <
|
|
opName, (outs SReg_64:$sdst), (ins SReg_64:$src0),
|
|
"$sdst, $src0", pattern
|
|
>;
|
|
|
|
// 64-bit input, 32-bit output.
|
|
class SOP1_32_64 <string opName, list<dag> pattern=[]> : SOP1_Pseudo <
|
|
opName, (outs SReg_32:$sdst), (ins SSrc_b64:$src0),
|
|
"$sdst, $src0", pattern
|
|
>;
|
|
|
|
// 32-bit input, 64-bit output.
|
|
class SOP1_64_32 <string opName, list<dag> pattern=[], bit tied_in = 0> : SOP1_Pseudo <
|
|
opName, (outs SReg_64:$sdst),
|
|
!if(tied_in, (ins SSrc_b32:$src0, SReg_64:$sdst_in),
|
|
(ins SSrc_b32:$src0)),
|
|
"$sdst, $src0", pattern> {
|
|
let Constraints = !if(tied_in, "$sdst = $sdst_in", "");
|
|
}
|
|
|
|
// no input, 64-bit output.
|
|
class SOP1_64_0 <string opName, list<dag> pattern=[]> : SOP1_Pseudo <
|
|
opName, (outs SReg_64:$sdst), (ins), "$sdst", pattern> {
|
|
let has_src0 = 0;
|
|
}
|
|
|
|
// 64-bit input, no output
|
|
class SOP1_1 <string opName, list<dag> pattern=[]> : SOP1_Pseudo <
|
|
opName, (outs), (ins SReg_64:$src0), "$src0", pattern> {
|
|
let has_sdst = 0;
|
|
}
|
|
|
|
class SOP1_1_REGIMM64 <string opName, list<dag> pattern=[]> : SOP1_Pseudo <
|
|
opName, (outs), (ins SSrc_b64:$src0), "$src0", pattern> {
|
|
let has_sdst = 0;
|
|
}
|
|
|
|
class UniformUnaryFrag<SDPatternOperator Op> : PatFrag <
|
|
(ops node:$src0),
|
|
(Op $src0),
|
|
[{ return !N->isDivergent(); }]> {
|
|
// This check is unnecessary as it's captured by the result register
|
|
// bank constraint.
|
|
//
|
|
// FIXME: Should add a way for the emitter to recognize this is a
|
|
// trivially true predicate to eliminate the check.
|
|
let GISelPredicateCode = [{return true;}];
|
|
}
|
|
|
|
class UniformBinFrag<SDPatternOperator Op> : PatFrag <
|
|
(ops node:$src0, node:$src1),
|
|
(Op $src0, $src1),
|
|
[{ return !N->isDivergent(); }]> {
|
|
// This check is unnecessary as it's captured by the result register
|
|
// bank constraint.
|
|
//
|
|
// FIXME: Should add a way for the emitter to recognize this is a
|
|
// trivially true predicate to eliminate the check.
|
|
let GISelPredicateCode = [{return true;}];
|
|
}
|
|
|
|
class UniformTernaryFrag<SDPatternOperator Op> : PatFrag <
|
|
(ops node:$src0, node:$src1, node:$src2),
|
|
(Op $src0, $src1, $src2),
|
|
[{ return !N->isDivergent(); }]> {
|
|
// This check is unnecessary as it's captured by the result register
|
|
// bank constraint.
|
|
//
|
|
// FIXME: Should add a way for the emitter to recognize this is a
|
|
// trivially true predicate to eliminate the check.
|
|
let GISelPredicateCode = [{return true;}];
|
|
}
|
|
|
|
class DivergentBinFrag<SDPatternOperator Op> : PatFrag <
|
|
(ops node:$src0, node:$src1),
|
|
(Op $src0, $src1),
|
|
[{ return N->isDivergent(); }]> {
|
|
// This check is unnecessary as it's captured by the result register
|
|
// bank constraint.
|
|
//
|
|
// FIXME: Should add a way for the emitter to recognize this is a
|
|
// trivially true predicate to eliminate the check.
|
|
let GISelPredicateCode = [{return true;}];
|
|
}
|
|
|
|
|
|
let isMoveImm = 1 in {
|
|
let isReMaterializable = 1, isAsCheapAsAMove = 1 in {
|
|
def S_MOV_B32 : SOP1_32 <"s_mov_b32">;
|
|
def S_MOV_B64 : SOP1_64 <"s_mov_b64">;
|
|
} // End isReMaterializable = 1
|
|
|
|
let Uses = [SCC] in {
|
|
def S_CMOV_B32 : SOP1_32 <"s_cmov_b32">;
|
|
def S_CMOV_B64 : SOP1_64 <"s_cmov_b64">;
|
|
} // End Uses = [SCC]
|
|
} // End isMoveImm = 1
|
|
|
|
// Variant of S_MOV_B32 used for reading from volatile registers like
|
|
// SRC_POPS_EXITING_WAVE_ID.
|
|
let hasSideEffects = 1 in
|
|
def S_MOV_B32_sideeffects : SOP1_32 <"s_mov_b32">;
|
|
|
|
let Defs = [SCC] in {
|
|
def S_NOT_B32 : SOP1_32 <"s_not_b32",
|
|
[(set i32:$sdst, (UniformUnaryFrag<not> i32:$src0))]
|
|
>;
|
|
|
|
def S_NOT_B64 : SOP1_64 <"s_not_b64",
|
|
[(set i64:$sdst, (UniformUnaryFrag<not> i64:$src0))]
|
|
>;
|
|
def S_WQM_B32 : SOP1_32 <"s_wqm_b32",
|
|
[(set i32:$sdst, (int_amdgcn_s_wqm i32:$src0))]>;
|
|
def S_WQM_B64 : SOP1_64 <"s_wqm_b64",
|
|
[(set i64:$sdst, (int_amdgcn_s_wqm i64:$src0))]>;
|
|
} // End Defs = [SCC]
|
|
|
|
|
|
let WaveSizePredicate = isWave32 in {
|
|
def : GCNPat <
|
|
(int_amdgcn_wqm_vote i1:$src0),
|
|
(S_WQM_B32 SSrc_b32:$src0)
|
|
>;
|
|
}
|
|
|
|
let WaveSizePredicate = isWave64 in {
|
|
def : GCNPat <
|
|
(int_amdgcn_wqm_vote i1:$src0),
|
|
(S_WQM_B64 SSrc_b64:$src0)
|
|
>;
|
|
}
|
|
|
|
let isReMaterializable = 1, isAsCheapAsAMove = 1 in {
|
|
def S_BREV_B32 : SOP1_32 <"s_brev_b32",
|
|
[(set i32:$sdst, (UniformUnaryFrag<bitreverse> i32:$src0))]
|
|
>;
|
|
def S_BREV_B64 : SOP1_64 <"s_brev_b64",
|
|
[(set i64:$sdst, (UniformUnaryFrag<bitreverse> i64:$src0))]
|
|
>;
|
|
} // End isReMaterializable = 1, isAsCheapAsAMove = 1
|
|
|
|
let Defs = [SCC] in {
|
|
def S_BCNT0_I32_B32 : SOP1_32 <"s_bcnt0_i32_b32">;
|
|
def S_BCNT0_I32_B64 : SOP1_32_64 <"s_bcnt0_i32_b64">;
|
|
def S_BCNT1_I32_B32 : SOP1_32 <"s_bcnt1_i32_b32",
|
|
[(set i32:$sdst, (UniformUnaryFrag<ctpop> i32:$src0))]
|
|
>;
|
|
def S_BCNT1_I32_B64 : SOP1_32_64 <"s_bcnt1_i32_b64",
|
|
[(set i32:$sdst, (UniformUnaryFrag<ctpop> i64:$src0))]
|
|
>;
|
|
} // End Defs = [SCC]
|
|
|
|
let isReMaterializable = 1 in {
|
|
def S_FF0_I32_B32 : SOP1_32 <"s_ff0_i32_b32">;
|
|
def S_FF0_I32_B64 : SOP1_32_64 <"s_ff0_i32_b64">;
|
|
def S_FF1_I32_B64 : SOP1_32_64 <"s_ff1_i32_b64",
|
|
[(set i32:$sdst, (UniformUnaryFrag<AMDGPUffbl_b32> i64:$src0))]
|
|
>;
|
|
|
|
def S_FF1_I32_B32 : SOP1_32 <"s_ff1_i32_b32",
|
|
[(set i32:$sdst, (UniformUnaryFrag<AMDGPUffbl_b32> i32:$src0))]
|
|
>;
|
|
|
|
def S_FLBIT_I32_B32 : SOP1_32 <"s_flbit_i32_b32",
|
|
[(set i32:$sdst, (UniformUnaryFrag<AMDGPUffbh_u32> i32:$src0))]
|
|
>;
|
|
|
|
def S_FLBIT_I32_B64 : SOP1_32_64 <"s_flbit_i32_b64",
|
|
[(set i32:$sdst, (UniformUnaryFrag<AMDGPUffbh_u32> i64:$src0))]
|
|
>;
|
|
def S_FLBIT_I32 : SOP1_32 <"s_flbit_i32",
|
|
[(set i32:$sdst, (UniformUnaryFrag<int_amdgcn_sffbh> i32:$src0))]
|
|
>;
|
|
def S_FLBIT_I32_I64 : SOP1_32_64 <"s_flbit_i32_i64">;
|
|
def S_SEXT_I32_I8 : SOP1_32 <"s_sext_i32_i8",
|
|
[(set i32:$sdst, (UniformSextInreg<i8> i32:$src0))]
|
|
>;
|
|
def S_SEXT_I32_I16 : SOP1_32 <"s_sext_i32_i16",
|
|
[(set i32:$sdst, (UniformSextInreg<i16> i32:$src0))]
|
|
>;
|
|
} // End isReMaterializable = 1
|
|
|
|
def S_BITSET0_B32 : SOP1_32 <"s_bitset0_b32", [], 1>;
|
|
def S_BITSET0_B64 : SOP1_64_32 <"s_bitset0_b64", [], 1>;
|
|
def S_BITSET1_B32 : SOP1_32 <"s_bitset1_b32", [], 1>;
|
|
def S_BITSET1_B64 : SOP1_64_32 <"s_bitset1_b64", [], 1>;
|
|
|
|
def S_GETPC_B64 : SOP1_64_0 <"s_getpc_b64">;
|
|
// PSEUDO includes a workaround for a hardware anomaly where some ASICs
|
|
// zero-extend the result from 48 bits instead of sign-extending.
|
|
let isReMaterializable = 1 in
|
|
def S_GETPC_B64_pseudo : SOP1_64_0 <"s_getpc_b64",
|
|
[(set i64:$sdst, (int_amdgcn_s_getpc))]
|
|
>;
|
|
|
|
let isTerminator = 1, isBarrier = 1, SchedRW = [WriteBranch] in {
|
|
|
|
let isBranch = 1, isIndirectBranch = 1 in {
|
|
def S_SETPC_B64 : SOP1_1 <"s_setpc_b64">;
|
|
|
|
let SubtargetPredicate = HasAddPC64Inst in
|
|
def S_ADD_PC_I64 : SOP1_1_REGIMM64 <"s_add_pc_i64">;
|
|
} // End isBranch = 1, isIndirectBranch = 1
|
|
|
|
let isReturn = 1 in {
|
|
// Define variant marked as return rather than branch.
|
|
def S_SETPC_B64_return : SOP1_1<"">;
|
|
}
|
|
} // End isTerminator = 1, isBarrier = 1
|
|
|
|
let isCall = 1 in {
|
|
def S_SWAPPC_B64 : SOP1_64 <"s_swappc_b64"
|
|
>;
|
|
}
|
|
|
|
def S_RFE_B64 : SOP1_1 <"s_rfe_b64">;
|
|
|
|
let hasSideEffects = 1, Uses = [EXEC], Defs = [EXEC, SCC] in {
|
|
|
|
def S_AND_SAVEEXEC_B64 : SOP1_64 <"s_and_saveexec_b64">;
|
|
def S_OR_SAVEEXEC_B64 : SOP1_64 <"s_or_saveexec_b64">;
|
|
def S_XOR_SAVEEXEC_B64 : SOP1_64 <"s_xor_saveexec_b64">;
|
|
def S_ANDN2_SAVEEXEC_B64 : SOP1_64 <"s_andn2_saveexec_b64">;
|
|
def S_ORN2_SAVEEXEC_B64 : SOP1_64 <"s_orn2_saveexec_b64">;
|
|
def S_NAND_SAVEEXEC_B64 : SOP1_64 <"s_nand_saveexec_b64">;
|
|
def S_NOR_SAVEEXEC_B64 : SOP1_64 <"s_nor_saveexec_b64">;
|
|
def S_XNOR_SAVEEXEC_B64 : SOP1_64 <"s_xnor_saveexec_b64">;
|
|
|
|
} // End hasSideEffects = 1, Uses = [EXEC], Defs = [EXEC, SCC]
|
|
|
|
let Defs = [SCC] in {
|
|
def S_QUADMASK_B32 : SOP1_32 <"s_quadmask_b32",
|
|
[(set i32:$sdst, (int_amdgcn_s_quadmask i32:$src0))]>;
|
|
def S_QUADMASK_B64 : SOP1_64 <"s_quadmask_b64",
|
|
[(set i64:$sdst, (int_amdgcn_s_quadmask i64:$src0))]>;
|
|
}
|
|
|
|
let Uses = [M0] in {
|
|
def S_MOVRELS_B32 : SOP1_32R <"s_movrels_b32">;
|
|
def S_MOVRELS_B64 : SOP1_64R <"s_movrels_b64">;
|
|
def S_MOVRELD_B32 : SOP1_32_movreld <"s_movreld_b32">;
|
|
def S_MOVRELD_B64 : SOP1_64_movreld <"s_movreld_b64">;
|
|
} // End Uses = [M0]
|
|
|
|
let SubtargetPredicate = isGFX6GFX7GFX8GFX9 in {
|
|
def S_CBRANCH_JOIN : SOP1_0_32R <"s_cbranch_join">;
|
|
} // End SubtargetPredicate = isGFX6GFX7GFX8GFX9
|
|
|
|
let Defs = [SCC] in {
|
|
def S_ABS_I32 : SOP1_32 <"s_abs_i32",
|
|
[(set i32:$sdst, (UniformUnaryFrag<abs> i32:$src0))]
|
|
>;
|
|
} // End Defs = [SCC]
|
|
|
|
let SubtargetPredicate = HasVGPRIndexMode in {
|
|
def S_SET_GPR_IDX_IDX : SOP1_0_32<"s_set_gpr_idx_idx"> {
|
|
let Uses = [M0, MODE];
|
|
let Defs = [M0, MODE];
|
|
}
|
|
}
|
|
|
|
let SubtargetPredicate = isGFX9Plus in {
|
|
let hasSideEffects = 1, Defs = [EXEC, SCC], Uses = [EXEC] in {
|
|
def S_ANDN1_SAVEEXEC_B64 : SOP1_64<"s_andn1_saveexec_b64">;
|
|
def S_ORN1_SAVEEXEC_B64 : SOP1_64<"s_orn1_saveexec_b64">;
|
|
def S_ANDN1_WREXEC_B64 : SOP1_64<"s_andn1_wrexec_b64">;
|
|
def S_ANDN2_WREXEC_B64 : SOP1_64<"s_andn2_wrexec_b64">;
|
|
} // End hasSideEffects = 1, Defs = [EXEC, SCC], Uses = [EXEC]
|
|
|
|
let isReMaterializable = 1 in
|
|
def S_BITREPLICATE_B64_B32 : SOP1_64_32<"s_bitreplicate_b64_b32",
|
|
[(set i64:$sdst, (int_amdgcn_s_bitreplicate i32:$src0))]>;
|
|
} // End SubtargetPredicate = isGFX9Plus
|
|
|
|
let SubtargetPredicate = isGFX10Plus in {
|
|
let hasSideEffects = 1, Defs = [EXEC, SCC], Uses = [EXEC] in {
|
|
def S_AND_SAVEEXEC_B32 : SOP1_32<"s_and_saveexec_b32">;
|
|
def S_OR_SAVEEXEC_B32 : SOP1_32<"s_or_saveexec_b32">;
|
|
def S_XOR_SAVEEXEC_B32 : SOP1_32<"s_xor_saveexec_b32">;
|
|
def S_ANDN2_SAVEEXEC_B32 : SOP1_32<"s_andn2_saveexec_b32">;
|
|
def S_ORN2_SAVEEXEC_B32 : SOP1_32<"s_orn2_saveexec_b32">;
|
|
def S_NAND_SAVEEXEC_B32 : SOP1_32<"s_nand_saveexec_b32">;
|
|
def S_NOR_SAVEEXEC_B32 : SOP1_32<"s_nor_saveexec_b32">;
|
|
def S_XNOR_SAVEEXEC_B32 : SOP1_32<"s_xnor_saveexec_b32">;
|
|
def S_ANDN1_SAVEEXEC_B32 : SOP1_32<"s_andn1_saveexec_b32">;
|
|
def S_ORN1_SAVEEXEC_B32 : SOP1_32<"s_orn1_saveexec_b32">;
|
|
def S_ANDN1_WREXEC_B32 : SOP1_32<"s_andn1_wrexec_b32">;
|
|
def S_ANDN2_WREXEC_B32 : SOP1_32<"s_andn2_wrexec_b32">;
|
|
} // End hasSideEffects = 1, Defs = [EXEC, SCC], Uses = [EXEC]
|
|
|
|
let Uses = [M0] in {
|
|
def S_MOVRELSD_2_B32 : SOP1_32<"s_movrelsd_2_b32">;
|
|
} // End Uses = [M0]
|
|
} // End SubtargetPredicate = isGFX10Plus
|
|
|
|
let SubtargetPredicate = isGFX11Plus in {
|
|
let hasSideEffects = 1 in {
|
|
// For s_sendmsg_rtn_* the src0 field encodes the message type directly; it
|
|
// is not an SGPR number.
|
|
def S_SENDMSG_RTN_B32 : SOP1_Pseudo<
|
|
"s_sendmsg_rtn_b32", (outs SReg_32:$sdst), (ins SendMsg:$src0),
|
|
"$sdst, $src0", [(set i32:$sdst, (int_amdgcn_s_sendmsg_rtn timm:$src0))]
|
|
>;
|
|
def S_SENDMSG_RTN_B64 : SOP1_Pseudo<
|
|
"s_sendmsg_rtn_b64", (outs SReg_64:$sdst), (ins SendMsg:$src0),
|
|
"$sdst, $src0", [(set i64:$sdst, (int_amdgcn_s_sendmsg_rtn timm:$src0))]
|
|
>;
|
|
}
|
|
} // End SubtargetPredicate = isGFX11Plus
|
|
|
|
let SubtargetPredicate = isGFX12Plus in {
|
|
let hasSideEffects = 1, isConvergent = 1, Defs = [SCC] in {
|
|
def S_ALLOC_VGPR : SOP1_0_32 <"s_alloc_vgpr",
|
|
[(set SCC, (int_amdgcn_s_alloc_vgpr SSrc_b32:$src0))]
|
|
>;
|
|
}
|
|
} // End SubtargetPredicate = isGFX12Plus
|
|
|
|
class SOP1_F32_Inst<string opName, SDPatternOperator Op, ValueType vt0=f32,
|
|
ValueType vt1=vt0> :
|
|
SOP1_32<opName, [(set vt0:$sdst, (UniformUnaryFrag<Op> vt1:$src0))]>;
|
|
|
|
let SubtargetPredicate = HasSALUFloatInsts, Uses = [MODE],
|
|
SchedRW = [WriteSFPU], isReMaterializable = 1 in {
|
|
def S_CVT_F32_I32 : SOP1_F32_Inst<"s_cvt_f32_i32", sint_to_fp, f32, i32>;
|
|
def S_CVT_F32_U32 : SOP1_F32_Inst<"s_cvt_f32_u32", uint_to_fp, f32, i32>;
|
|
|
|
let mayRaiseFPException = 1 in {
|
|
def S_CVT_I32_F32 : SOP1_F32_Inst<"s_cvt_i32_f32", fp_to_sint, i32, f32>;
|
|
def S_CVT_U32_F32 : SOP1_F32_Inst<"s_cvt_u32_f32", fp_to_uint, i32, f32>;
|
|
def S_CVT_F32_F16 : SOP1_F32_Inst<"s_cvt_f32_f16", fpextend, f32, f16>;
|
|
def S_CVT_HI_F32_F16 : SOP1_32<"s_cvt_hi_f32_f16">;
|
|
|
|
def S_CEIL_F32 : SOP1_F32_Inst<"s_ceil_f32", fceil>;
|
|
def S_FLOOR_F32 : SOP1_F32_Inst<"s_floor_f32", ffloor>;
|
|
def S_TRUNC_F32 : SOP1_F32_Inst<"s_trunc_f32", ftrunc>;
|
|
def S_RNDNE_F32 : SOP1_F32_Inst<"s_rndne_f32", froundeven>;
|
|
|
|
let FPDPRounding = 1 in
|
|
def S_CVT_F16_F32 : SOP1_F32_Inst<"s_cvt_f16_f32", fpround, f16, f32>;
|
|
|
|
def S_CEIL_F16 : SOP1_F32_Inst<"s_ceil_f16", fceil, f16>;
|
|
def S_FLOOR_F16 : SOP1_F32_Inst<"s_floor_f16", ffloor, f16>;
|
|
def S_TRUNC_F16 : SOP1_F32_Inst<"s_trunc_f16", ftrunc, f16>;
|
|
def S_RNDNE_F16 : SOP1_F32_Inst<"s_rndne_f16", froundeven, f16>;
|
|
} // End mayRaiseFPException = 1
|
|
} // End SubtargetPredicate = HasSALUFloatInsts, Uses = [MODE]
|
|
// SchedRW = [WriteSFPU], isReMaterializable = 1
|
|
|
|
let SubtargetPredicate = HasSALUFloatInsts, AddedComplexity = 9 in {
|
|
// Fallback patterns for f32->i16 conversion.
|
|
def : GCNPat<(i16 (UniformUnaryFrag<fp_to_sint> f32:$src0)),
|
|
(S_CVT_I32_F32 $src0)>;
|
|
def : GCNPat<(i16 (UniformUnaryFrag<fp_to_uint> f32:$src0)),
|
|
(S_CVT_U32_F32 $src0)>;
|
|
// f16 -> i32 : form chain f16 -> f32 -> i32
|
|
def : GCNPat<(i32 (UniformUnaryFrag<fp_to_sint> f16:$src0)),
|
|
(S_CVT_I32_F32 (S_CVT_F32_F16 $src0))>;
|
|
def : GCNPat<(i32 (UniformUnaryFrag<fp_to_uint> f16:$src0)),
|
|
(S_CVT_U32_F32 (S_CVT_F32_F16 $src0))>;
|
|
|
|
// i32 -> f16 : form chain i32 -> f32 -> f16
|
|
def : GCNPat<(f16 (UniformUnaryFrag<sint_to_fp> i32:$src0)),
|
|
(S_CVT_F16_F32 (S_CVT_F32_I32 $src0))>;
|
|
def : GCNPat<(f16 (UniformUnaryFrag<uint_to_fp> i32:$src0)),
|
|
(S_CVT_F16_F32 (S_CVT_F32_U32 $src0))>;
|
|
|
|
// f32 -> i32 saturated conversions
|
|
def : GCNPat<(i32 (UniformBinFrag<fp_to_sint_sat> f32:$src0, i32)),
|
|
(S_CVT_I32_F32 (f32 $src0))>;
|
|
def : GCNPat<(i32 (UniformBinFrag<fp_to_uint_sat> f32:$src0, i32)),
|
|
(S_CVT_U32_F32 (f32 $src0))>;
|
|
|
|
def : GCNPat<(i32 (UniformUnaryFrag<fp_to_sint_sat_gi> f32:$src0)),
|
|
(S_CVT_I32_F32 (f32 $src0))>;
|
|
def : GCNPat<(i32 (UniformUnaryFrag<fp_to_uint_sat_gi> f32:$src0)),
|
|
(S_CVT_U32_F32 (f32 $src0))>;
|
|
}
|
|
|
|
let hasSideEffects = 1 in {
|
|
let has_sdst = 0 in {
|
|
let Uses = [M0] in {
|
|
def S_BARRIER_SIGNAL_M0 : SOP1_Pseudo <"s_barrier_signal m0", (outs), (ins),
|
|
"", []>{
|
|
let SchedRW = [WriteBarrier];
|
|
let isConvergent = 1;
|
|
}
|
|
|
|
def S_BARRIER_SIGNAL_ISFIRST_M0 : SOP1_Pseudo <"s_barrier_signal_isfirst m0", (outs), (ins),
|
|
"", []>{
|
|
let Defs = [SCC];
|
|
let Uses = [M0, SCC];
|
|
let SchedRW = [WriteBarrier];
|
|
let isConvergent = 1;
|
|
}
|
|
|
|
def S_BARRIER_INIT_M0 : SOP1_Pseudo <"s_barrier_init m0", (outs), (ins),
|
|
"", []>{
|
|
let SchedRW = [WriteBarrier];
|
|
let isConvergent = 1;
|
|
}
|
|
|
|
def S_BARRIER_INIT_IMM : SOP1_Pseudo <"s_barrier_init", (outs),
|
|
(ins SplitBarrier:$src0), "$src0", []>{
|
|
let SchedRW = [WriteBarrier];
|
|
let isConvergent = 1;
|
|
}
|
|
|
|
def S_BARRIER_JOIN_M0 : SOP1_Pseudo <"s_barrier_join m0", (outs), (ins),
|
|
"", []>{
|
|
let SchedRW = [WriteBarrier];
|
|
let isConvergent = 1;
|
|
}
|
|
|
|
def S_WAKEUP_BARRIER_M0 : SOP1_Pseudo <"s_wakeup_barrier m0", (outs), (ins),
|
|
"", []>{
|
|
let SchedRW = [WriteBarrier];
|
|
let isConvergent = 1;
|
|
let SubtargetPredicate = HasSWakeupBarrier;
|
|
}
|
|
} // End Uses = [M0]
|
|
|
|
def S_BARRIER_SIGNAL_IMM : SOP1_Pseudo <"s_barrier_signal", (outs),
|
|
(ins SplitBarrier:$src0), "$src0", [(int_amdgcn_s_barrier_signal timm:$src0)]>{
|
|
let SchedRW = [WriteBarrier];
|
|
let isConvergent = 1;
|
|
}
|
|
|
|
def S_BARRIER_SIGNAL_ISFIRST_IMM : SOP1_Pseudo <"s_barrier_signal_isfirst", (outs),
|
|
(ins SplitBarrier:$src0), "$src0", [(set SCC, (int_amdgcn_s_barrier_signal_isfirst timm:$src0))]>{
|
|
let Defs = [SCC];
|
|
let Uses = [SCC];
|
|
let usesCustomInserter = 1;
|
|
let SchedRW = [WriteBarrier];
|
|
let isConvergent = 1;
|
|
}
|
|
|
|
def S_BARRIER_JOIN_IMM : SOP1_Pseudo <"s_barrier_join", (outs),
|
|
(ins SplitBarrier:$src0), "$src0", []>{
|
|
let SchedRW = [WriteBarrier];
|
|
let isConvergent = 1;
|
|
}
|
|
|
|
def S_WAKEUP_BARRIER_IMM : SOP1_Pseudo <"s_wakeup_barrier", (outs),
|
|
(ins SplitBarrier:$src0), "$src0", []>{
|
|
let SchedRW = [WriteBarrier];
|
|
let isConvergent = 1;
|
|
let SubtargetPredicate = HasSWakeupBarrier;
|
|
}
|
|
} // End has_sdst = 0
|
|
|
|
def S_GET_BARRIER_STATE_IMM : SOP1_Pseudo <"s_get_barrier_state", (outs SSrc_b32:$sdst),
|
|
(ins SplitBarrier:$src0), "$sdst, $src0", []>{
|
|
let SchedRW = [WriteBarrier];
|
|
let isConvergent = 1;
|
|
}
|
|
|
|
def S_GET_BARRIER_STATE_M0 : SOP1_Pseudo <"s_get_barrier_state $sdst, m0", (outs SSrc_b32:$sdst),
|
|
(ins), "", []>{
|
|
let Uses = [M0];
|
|
let SchedRW = [WriteBarrier];
|
|
let isConvergent = 1;
|
|
}
|
|
} // End hasSideEffects = 1
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOP2 Instructions
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
class SOP2_Pseudo<string opName, dag outs, dag ins,
|
|
string asmOps, list<dag> pattern=[]> :
|
|
SOP_Pseudo<opName, outs, ins, " " # asmOps, pattern> {
|
|
|
|
let mayLoad = 0;
|
|
let mayStore = 0;
|
|
let hasSideEffects = 0;
|
|
let SALU = 1;
|
|
let SOP2 = 1;
|
|
let SchedRW = [WriteSALU];
|
|
let UseNamedOperandTable = 1;
|
|
|
|
let has_sdst = 1;
|
|
|
|
// Pseudo instructions have no encodings, but adding this field here allows
|
|
// us to do:
|
|
// let sdst = xxx in {
|
|
// for multiclasses that include both real and pseudo instructions.
|
|
// field bits<7> sdst = 0;
|
|
}
|
|
|
|
class SOP2_Real<SOP_Pseudo ps, string name = ps.Mnemonic> :
|
|
InstSI <ps.OutOperandList, ps.InOperandList,
|
|
name # ps.AsmOperands> {
|
|
let SALU = 1;
|
|
let SOP2 = 1;
|
|
let isPseudo = 0;
|
|
let isCodeGenOnly = 0;
|
|
|
|
// copy relevant pseudo op flags
|
|
let SubtargetPredicate = ps.SubtargetPredicate;
|
|
let AsmMatchConverter = ps.AsmMatchConverter;
|
|
let UseNamedOperandTable = ps.UseNamedOperandTable;
|
|
let TSFlags = ps.TSFlags;
|
|
let SchedRW = ps.SchedRW;
|
|
let mayLoad = ps.mayLoad;
|
|
let mayStore = ps.mayStore;
|
|
let Constraints = ps.Constraints;
|
|
let Uses = ps.Uses;
|
|
let Defs = ps.Defs;
|
|
let isConvergent = ps.isConvergent;
|
|
|
|
// encoding
|
|
bits<7> sdst;
|
|
bits<8> src0;
|
|
bits<8> src1;
|
|
bits<32> imm;
|
|
}
|
|
|
|
class SOP2_Real32<bits<7> op, SOP_Pseudo ps, string name = ps.Mnemonic> :
|
|
SOP2_Real<ps, name>, Enc32 {
|
|
let Inst{7-0} = src0;
|
|
let Inst{15-8} = src1;
|
|
let Inst{22-16} = !if(ps.has_sdst, sdst, ?);
|
|
let Inst{29-23} = op;
|
|
let Inst{31-30} = 0x2; // encoding
|
|
}
|
|
|
|
class SOP2_Real64<bits<7> op, SOP_Pseudo ps, string name = ps.Mnemonic> :
|
|
SOP2_Real<ps, name>, Enc64 {
|
|
let Inst{7-0} = src0;
|
|
let Inst{15-8} = src1;
|
|
let Inst{22-16} = !if(ps.has_sdst, sdst, ?);
|
|
let Inst{29-23} = op;
|
|
let Inst{31-30} = 0x2; // encoding
|
|
let Inst{63-32} = imm;
|
|
}
|
|
|
|
class SOP2_F16 <string opName, list<dag> pattern=[]> : SOP2_Pseudo <
|
|
opName, (outs SReg_32:$sdst), (ins SSrc_f16:$src0, SSrc_f16:$src1),
|
|
"$sdst, $src0, $src1", pattern
|
|
>;
|
|
|
|
class SOP2_32 <string opName, list<dag> pattern=[]> : SOP2_Pseudo <
|
|
opName, (outs SReg_32:$sdst), (ins SSrc_b32:$src0, SSrc_b32:$src1),
|
|
"$sdst, $src0, $src1", pattern
|
|
>;
|
|
|
|
class SOP2_F32 <string opName, list<dag> pattern=[]> : SOP2_Pseudo <
|
|
opName, (outs SReg_32:$sdst), (ins SSrc_f32:$src0, SSrc_f32:$src1),
|
|
"$sdst, $src0, $src1", pattern
|
|
>;
|
|
|
|
class SOP2_64 <string opName, list<dag> pattern=[]> : SOP2_Pseudo <
|
|
opName, (outs SReg_64:$sdst), (ins SSrc_b64:$src0, SSrc_b64:$src1),
|
|
"$sdst, $src0, $src1", pattern
|
|
>;
|
|
|
|
class SOP2_64_32 <string opName, list<dag> pattern=[]> : SOP2_Pseudo <
|
|
opName, (outs SReg_64:$sdst), (ins SSrc_b64:$src0, SSrc_b32:$src1),
|
|
"$sdst, $src0, $src1", pattern
|
|
>;
|
|
|
|
class SOP2_64_32_32 <string opName, list<dag> pattern=[]> : SOP2_Pseudo <
|
|
opName, (outs SReg_64:$sdst), (ins SSrc_b32:$src0, SSrc_b32:$src1),
|
|
"$sdst, $src0, $src1", pattern
|
|
>;
|
|
|
|
|
|
let Defs = [SCC] in { // Carry out goes to SCC
|
|
let isCommutable = 1, isAdd = 1 in {
|
|
def S_ADD_U32 : SOP2_32 <"s_add_u32">;
|
|
def S_ADD_I32 : SOP2_32 <"s_add_i32",
|
|
[(set i32:$sdst, (UniformBinFrag<add> SSrc_b32:$src0, SSrc_b32:$src1))]
|
|
>;
|
|
} // End isCommutable = 1, isAdd = 1
|
|
|
|
def S_SUB_U32 : SOP2_32 <"s_sub_u32">;
|
|
def S_SUB_I32 : SOP2_32 <"s_sub_i32",
|
|
[(set i32:$sdst, (UniformBinFrag<sub> SSrc_b32:$src0, SSrc_b32:$src1))]
|
|
>;
|
|
|
|
let Uses = [SCC] in { // Carry in comes from SCC
|
|
let isCommutable = 1 in {
|
|
def S_ADDC_U32 : SOP2_32 <"s_addc_u32",
|
|
[(set i32:$sdst, (UniformBinFrag<adde> (i32 SSrc_b32:$src0), (i32 SSrc_b32:$src1)))]>;
|
|
} // End isCommutable = 1
|
|
|
|
def S_SUBB_U32 : SOP2_32 <"s_subb_u32",
|
|
[(set i32:$sdst, (UniformBinFrag<sube> (i32 SSrc_b32:$src0), (i32 SSrc_b32:$src1)))]>;
|
|
} // End Uses = [SCC]
|
|
|
|
let isCommutable = 1 in {
|
|
def S_MIN_I32 : SOP2_32 <"s_min_i32",
|
|
[(set i32:$sdst, (UniformBinFrag<smin> i32:$src0, i32:$src1))]
|
|
>;
|
|
def S_MIN_U32 : SOP2_32 <"s_min_u32",
|
|
[(set i32:$sdst, (UniformBinFrag<umin> i32:$src0, i32:$src1))]
|
|
>;
|
|
def S_MAX_I32 : SOP2_32 <"s_max_i32",
|
|
[(set i32:$sdst, (UniformBinFrag<smax> i32:$src0, i32:$src1))]
|
|
>;
|
|
def S_MAX_U32 : SOP2_32 <"s_max_u32",
|
|
[(set i32:$sdst, (UniformBinFrag<umax> i32:$src0, i32:$src1))]
|
|
>;
|
|
} // End isCommutable = 1
|
|
} // End Defs = [SCC]
|
|
|
|
let SubtargetPredicate = isGFX12Plus in {
|
|
def S_ADD_U64 : SOP2_64<"s_add_u64">{
|
|
let isCommutable = 1;
|
|
}
|
|
|
|
def S_SUB_U64 : SOP2_64<"s_sub_u64">;
|
|
|
|
def S_MUL_U64 : SOP2_64 <"s_mul_u64",
|
|
[(set i64:$sdst, (UniformBinFrag<mul> i64:$src0, i64:$src1))]> {
|
|
let isCommutable = 1;
|
|
}
|
|
|
|
// The higher 32-bits of the inputs contain the sign extension bits.
|
|
let hasSideEffects = 0 in {
|
|
def S_MUL_I64_I32_PSEUDO : SPseudoInstSI <
|
|
(outs SReg_64:$sdst), (ins SSrc_b64:$src0, SSrc_b64:$src1)
|
|
>;
|
|
|
|
// The higher 32-bits of the inputs are zero.
|
|
def S_MUL_U64_U32_PSEUDO : SPseudoInstSI <
|
|
(outs SReg_64:$sdst), (ins SSrc_b64:$src0, SSrc_b64:$src1)
|
|
>;
|
|
}
|
|
|
|
} // End SubtargetPredicate = isGFX12Plus
|
|
|
|
let Uses = [SCC] in {
|
|
def S_CSELECT_B32 : SOP2_32 <"s_cselect_b32">;
|
|
def S_CSELECT_B64 : SOP2_64 <"s_cselect_b64">;
|
|
} // End Uses = [SCC]
|
|
|
|
let Defs = [SCC] in {
|
|
let isCommutable = 1 in {
|
|
def S_AND_B32 : SOP2_32 <"s_and_b32",
|
|
[(set i32:$sdst, (UniformBinFrag<and> i32:$src0, i32:$src1))]
|
|
>;
|
|
|
|
def S_AND_B64 : SOP2_64 <"s_and_b64",
|
|
[(set i64:$sdst, (UniformBinFrag<and> i64:$src0, i64:$src1))]
|
|
>;
|
|
|
|
def S_OR_B32 : SOP2_32 <"s_or_b32",
|
|
[(set i32:$sdst, (UniformBinFrag<or> i32:$src0, i32:$src1))]
|
|
>;
|
|
|
|
def S_OR_B64 : SOP2_64 <"s_or_b64",
|
|
[(set i64:$sdst, (UniformBinFrag<or> i64:$src0, i64:$src1))]
|
|
>;
|
|
|
|
def S_XOR_B32 : SOP2_32 <"s_xor_b32",
|
|
[(set i32:$sdst, (UniformBinFrag<xor> i32:$src0, i32:$src1))]
|
|
>;
|
|
|
|
def S_XOR_B64 : SOP2_64 <"s_xor_b64",
|
|
[(set i64:$sdst, (UniformBinFrag<xor> i64:$src0, i64:$src1))]
|
|
>;
|
|
|
|
def S_XNOR_B32 : SOP2_32 <"s_xnor_b32",
|
|
[(set i32:$sdst, (UniformUnaryFrag<not> (xor_oneuse i32:$src0, i32:$src1)))]
|
|
>;
|
|
|
|
def S_XNOR_B64 : SOP2_64 <"s_xnor_b64",
|
|
[(set i64:$sdst, (UniformUnaryFrag<not> (xor_oneuse i64:$src0, i64:$src1)))]
|
|
>;
|
|
|
|
def S_NAND_B32 : SOP2_32 <"s_nand_b32",
|
|
[(set i32:$sdst, (UniformUnaryFrag<not> (and_oneuse i32:$src0, i32:$src1)))]
|
|
>;
|
|
|
|
def S_NAND_B64 : SOP2_64 <"s_nand_b64",
|
|
[(set i64:$sdst, (UniformUnaryFrag<not> (and_oneuse i64:$src0, i64:$src1)))]
|
|
>;
|
|
|
|
def S_NOR_B32 : SOP2_32 <"s_nor_b32",
|
|
[(set i32:$sdst, (UniformUnaryFrag<not> (or_oneuse i32:$src0, i32:$src1)))]
|
|
>;
|
|
|
|
def S_NOR_B64 : SOP2_64 <"s_nor_b64",
|
|
[(set i64:$sdst, (UniformUnaryFrag<not> (or_oneuse i64:$src0, i64:$src1)))]
|
|
>;
|
|
} // End isCommutable = 1
|
|
|
|
// There are also separate patterns for types other than i32
|
|
def S_ANDN2_B32 : SOP2_32 <"s_andn2_b32",
|
|
[(set i32:$sdst, (UniformBinFrag<and> i32:$src0, (not i32:$src1)))]
|
|
>;
|
|
|
|
def S_ANDN2_B64 : SOP2_64 <"s_andn2_b64",
|
|
[(set i64:$sdst, (UniformBinFrag<and> i64:$src0, (not i64:$src1)))]
|
|
>;
|
|
|
|
def S_ORN2_B32 : SOP2_32 <"s_orn2_b32",
|
|
[(set i32:$sdst, (UniformBinFrag<or> i32:$src0, (not i32:$src1)))]
|
|
>;
|
|
|
|
def S_ORN2_B64 : SOP2_64 <"s_orn2_b64",
|
|
[(set i64:$sdst, (UniformBinFrag<or> i64:$src0, (not i64:$src1)))]
|
|
>;
|
|
} // End Defs = [SCC]
|
|
|
|
// Use added complexity so these patterns are preferred to the VALU patterns.
|
|
let AddedComplexity = 1 in {
|
|
|
|
let Defs = [SCC] in {
|
|
// TODO: b64 versions require VOP3 change since v_lshlrev_b64 is VOP3
|
|
def S_LSHL_B32 : SOP2_32 <"s_lshl_b32",
|
|
[(set SReg_32:$sdst, (UniformBinFrag<cshl_32> (i32 SSrc_b32:$src0), (i32 SSrc_b32:$src1)))]
|
|
>;
|
|
def S_LSHL_B64 : SOP2_64_32 <"s_lshl_b64",
|
|
[(set SReg_64:$sdst, (UniformBinFrag<cshl_64> (i64 SSrc_b64:$src0), (i32 SSrc_b32:$src1)))]
|
|
>;
|
|
def S_LSHR_B32 : SOP2_32 <"s_lshr_b32",
|
|
[(set SReg_32:$sdst, (UniformBinFrag<csrl_32> (i32 SSrc_b32:$src0), (i32 SSrc_b32:$src1)))]
|
|
>;
|
|
def S_LSHR_B64 : SOP2_64_32 <"s_lshr_b64",
|
|
[(set SReg_64:$sdst, (UniformBinFrag<csrl_64> (i64 SSrc_b64:$src0), (i32 SSrc_b32:$src1)))]
|
|
>;
|
|
def S_ASHR_I32 : SOP2_32 <"s_ashr_i32",
|
|
[(set SReg_32:$sdst, (UniformBinFrag<csra_32> (i32 SSrc_b32:$src0), (i32 SSrc_b32:$src1)))]
|
|
>;
|
|
def S_ASHR_I64 : SOP2_64_32 <"s_ashr_i64",
|
|
[(set SReg_64:$sdst, (UniformBinFrag<csra_64> (i64 SSrc_b64:$src0), (i32 SSrc_b32:$src1)))]
|
|
>;
|
|
} // End Defs = [SCC]
|
|
|
|
let isReMaterializable = 1 in {
|
|
def S_BFM_B32 : SOP2_32 <"s_bfm_b32",
|
|
[(set i32:$sdst, (UniformBinFrag<AMDGPUbfm> i32:$src0, i32:$src1))]>;
|
|
def S_BFM_B64 : SOP2_64_32_32 <"s_bfm_b64">;
|
|
|
|
def S_MUL_I32 : SOP2_32 <"s_mul_i32",
|
|
[(set i32:$sdst, (UniformBinFrag<mul> i32:$src0, i32:$src1))]> {
|
|
let isCommutable = 1;
|
|
}
|
|
} // End isReMaterializable = 1
|
|
} // End AddedComplexity = 1
|
|
|
|
let Defs = [SCC] in {
|
|
def S_BFE_U32 : SOP2_32 <"s_bfe_u32">;
|
|
def S_BFE_I32 : SOP2_32 <"s_bfe_i32">;
|
|
def S_BFE_U64 : SOP2_64_32 <"s_bfe_u64">;
|
|
def S_BFE_I64 : SOP2_64_32 <"s_bfe_i64">;
|
|
} // End Defs = [SCC]
|
|
|
|
def S_CBRANCH_G_FORK : SOP2_Pseudo <
|
|
"s_cbranch_g_fork", (outs),
|
|
(ins SCSrc_b64:$src0, SCSrc_b64:$src1),
|
|
"$src0, $src1"
|
|
> {
|
|
let has_sdst = 0;
|
|
let SubtargetPredicate = isGFX6GFX7GFX8GFX9;
|
|
}
|
|
|
|
let isCommutable = 1, Defs = [SCC] in
|
|
def S_ABSDIFF_I32 : SOP2_32 <"s_absdiff_i32",
|
|
[(set i32:$sdst, (UniformUnaryFrag<abs> (sub_oneuse i32:$src0, i32:$src1)))]
|
|
>;
|
|
|
|
let SubtargetPredicate = isGFX8GFX9 in {
|
|
def S_RFE_RESTORE_B64 : SOP2_Pseudo <
|
|
"s_rfe_restore_b64", (outs),
|
|
(ins SSrc_b64:$src0, SSrc_b32:$src1),
|
|
"$src0, $src1"
|
|
> {
|
|
let hasSideEffects = 1;
|
|
let has_sdst = 0;
|
|
}
|
|
}
|
|
|
|
let SubtargetPredicate = isGFX9Plus in {
|
|
let isReMaterializable = 1 in {
|
|
def S_PACK_LL_B32_B16 : SOP2_32<"s_pack_ll_b32_b16">;
|
|
def S_PACK_LH_B32_B16 : SOP2_32<"s_pack_lh_b32_b16">;
|
|
def S_PACK_HH_B32_B16 : SOP2_32<"s_pack_hh_b32_b16">;
|
|
} // End isReMaterializable = 1
|
|
|
|
let Defs = [SCC] in {
|
|
def S_LSHL1_ADD_U32 : SOP2_32<"s_lshl1_add_u32",
|
|
[(set i32:$sdst, (shl1_add SSrc_b32:$src0, SSrc_b32:$src1))]
|
|
>;
|
|
def S_LSHL2_ADD_U32 : SOP2_32<"s_lshl2_add_u32",
|
|
[(set i32:$sdst, (shl2_add SSrc_b32:$src0, SSrc_b32:$src1))]
|
|
>;
|
|
def S_LSHL3_ADD_U32 : SOP2_32<"s_lshl3_add_u32",
|
|
[(set i32:$sdst, (shl3_add SSrc_b32:$src0, SSrc_b32:$src1))]
|
|
>;
|
|
def S_LSHL4_ADD_U32 : SOP2_32<"s_lshl4_add_u32",
|
|
[(set i32:$sdst, (shl4_add SSrc_b32:$src0, SSrc_b32:$src1))]
|
|
>;
|
|
} // End Defs = [SCC]
|
|
|
|
let isCommutable = 1, isReMaterializable = 1 in {
|
|
def S_MUL_HI_U32 : SOP2_32<"s_mul_hi_u32",
|
|
[(set i32:$sdst, (UniformBinFrag<mulhu> SSrc_b32:$src0, SSrc_b32:$src1))]>;
|
|
def S_MUL_HI_I32 : SOP2_32<"s_mul_hi_i32",
|
|
[(set i32:$sdst, (UniformBinFrag<mulhs> SSrc_b32:$src0, SSrc_b32:$src1))]>;
|
|
} // End isCommutable = 1, isReMaterializable = 1
|
|
} // End SubtargetPredicate = isGFX9Plus
|
|
|
|
let SubtargetPredicate = isGFX11Plus in {
|
|
def S_PACK_HL_B32_B16 : SOP2_32<"s_pack_hl_b32_b16">;
|
|
} // End SubtargetPredicate = isGFX11Plus
|
|
|
|
class SOP2_F32_Inst<string opName, SDPatternOperator Op, ValueType dstVt=f32> :
|
|
SOP2_F32<opName,
|
|
[(set dstVt:$sdst, (UniformBinFrag<Op> SSrc_f32:$src0, SSrc_f32:$src1))]>;
|
|
|
|
class SOP2_F16_Inst<string opName, SDPatternOperator Op> :
|
|
SOP2_F16<opName,
|
|
[(set f16:$sdst, (UniformBinFrag<Op> SSrc_f16:$src0, SSrc_f16:$src1))]>;
|
|
|
|
let SubtargetPredicate = HasSALUFloatInsts, mayRaiseFPException = 1,
|
|
Uses = [MODE], SchedRW = [WriteSFPU] in {
|
|
let isReMaterializable = 1 in {
|
|
let isCommutable = 1 in {
|
|
def S_ADD_F32 : SOP2_F32_Inst<"s_add_f32", any_fadd>;
|
|
def S_MIN_F32 : SOP2_F32_Inst<"s_min_f32", fminnum_like>;
|
|
def S_MAX_F32 : SOP2_F32_Inst<"s_max_f32", fmaxnum_like>;
|
|
def S_MUL_F32 : SOP2_F32_Inst<"s_mul_f32", any_fmul>;
|
|
|
|
let FixedSize = 1 in
|
|
def S_FMAAK_F32 : SOP2_Pseudo<
|
|
"s_fmaak_f32", (outs SReg_32:$sdst),
|
|
(ins SSrc_f32:$src0, SSrc_f32:$src1, KImmFP32:$imm),
|
|
"$sdst, $src0, $src1, $imm"
|
|
>;
|
|
|
|
let FPDPRounding = 1 in {
|
|
def S_ADD_F16 : SOP2_F16_Inst<"s_add_f16", any_fadd>;
|
|
def S_MUL_F16 : SOP2_F16_Inst<"s_mul_f16", any_fmul>;
|
|
} // End FPDPRounding
|
|
|
|
def S_MIN_F16 : SOP2_F16_Inst<"s_min_f16", fminnum_like>;
|
|
def S_MAX_F16 : SOP2_F16_Inst<"s_max_f16", fmaxnum_like>;
|
|
} // End isCommutable = 1
|
|
|
|
let FPDPRounding = 1 in
|
|
def S_SUB_F16 : SOP2_F16_Inst<"s_sub_f16", any_fsub>;
|
|
|
|
def S_SUB_F32 : SOP2_F32_Inst<"s_sub_f32", any_fsub>;
|
|
def S_CVT_PK_RTZ_F16_F32 : SOP2_F32_Inst<"s_cvt_pk_rtz_f16_f32",
|
|
AMDGPUpkrtz_f16_f32, v2f16>;
|
|
|
|
let FixedSize = 1 in
|
|
def S_FMAMK_F32 : SOP2_Pseudo<
|
|
"s_fmamk_f32", (outs SReg_32:$sdst),
|
|
(ins SSrc_f32:$src0, KImmFP32:$imm, SSrc_f32:$src1),
|
|
"$sdst, $src0, $imm, $src1"
|
|
>;
|
|
} // End isReMaterializable = 1
|
|
|
|
let Constraints = "$sdst = $src2",
|
|
isCommutable = 1, AddedComplexity = 20 in {
|
|
def S_FMAC_F32 : SOP2_Pseudo<
|
|
"s_fmac_f32", (outs SReg_32:$sdst),
|
|
(ins SSrc_f32:$src0, SSrc_f32:$src1, SReg_32:$src2),
|
|
"$sdst, $src0, $src1",
|
|
[(set f32:$sdst, (UniformTernaryFrag<any_fma> SSrc_f32:$src0, SSrc_f32:$src1, SReg_32:$src2))]
|
|
>;
|
|
|
|
def S_FMAC_F16 : SOP2_Pseudo<
|
|
"s_fmac_f16", (outs SReg_32:$sdst),
|
|
(ins SSrc_f16:$src0, SSrc_f16:$src1, SReg_32:$src2),
|
|
"$sdst, $src0, $src1",
|
|
[(set f16:$sdst, (UniformTernaryFrag<any_fma> SSrc_f16:$src0, SSrc_f16:$src1, SReg_32:$src2))]
|
|
>;
|
|
} // End Constraints = "$sdst = $src2",
|
|
// isCommutable = 1, AddedComplexity = 20
|
|
} // End SubtargetPredicate = HasSALUFloatInsts, mayRaiseFPException = 1,
|
|
// Uses = [MODE], SchedRW = [WriteSFPU]
|
|
|
|
// On GFX12 MIN/MAX instructions do not read MODE register.
|
|
let SubtargetPredicate = isGFX12Plus, mayRaiseFPException = 1, isCommutable = 1,
|
|
isReMaterializable = 1, SchedRW = [WriteSFPU], AddedComplexity = 25 in {
|
|
def S_MINIMUM_F32 : SOP2_F32_Inst<"s_minimum_f32", UniformBinFrag<fminimum>>;
|
|
def S_MAXIMUM_F32 : SOP2_F32_Inst<"s_maximum_f32", UniformBinFrag<fmaximum>>;
|
|
def S_MINIMUM_F16 : SOP2_F16_Inst<"s_minimum_f16", UniformBinFrag<fminimum>>;
|
|
def S_MAXIMUM_F16 : SOP2_F16_Inst<"s_maximum_f16", UniformBinFrag<fmaximum>>;
|
|
}
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOPK Instructions
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
class SOPK_Pseudo <string opName, dag outs, dag ins,
|
|
string asmOps, list<dag> pattern=[]> :
|
|
SOP_Pseudo<opName, outs, ins, " " # asmOps, pattern> {
|
|
let mayLoad = 0;
|
|
let mayStore = 0;
|
|
let hasSideEffects = 0;
|
|
let SALU = 1;
|
|
let SOPK = 1;
|
|
let FixedSize = 1;
|
|
let SchedRW = [WriteSALU];
|
|
let UseNamedOperandTable = 1;
|
|
|
|
let has_sdst = 1;
|
|
}
|
|
|
|
class SOPK_Real<SOPK_Pseudo ps, string name = ps.Mnemonic> :
|
|
InstSI <ps.OutOperandList, ps.InOperandList,
|
|
name # ps.AsmOperands> {
|
|
let SALU = 1;
|
|
let SOPK = 1;
|
|
let isPseudo = 0;
|
|
let isCodeGenOnly = 0;
|
|
let UseNamedOperandTable = 1;
|
|
|
|
// copy relevant pseudo op flags
|
|
let SubtargetPredicate = ps.SubtargetPredicate;
|
|
let AsmMatchConverter = ps.AsmMatchConverter;
|
|
let Constraints = ps.Constraints;
|
|
let SchedRW = ps.SchedRW;
|
|
let mayLoad = ps.mayLoad;
|
|
let mayStore = ps.mayStore;
|
|
let isBranch = ps.isBranch;
|
|
let isCall = ps.isCall;
|
|
let isTerminator = ps.isTerminator;
|
|
let isReturn = ps.isReturn;
|
|
let isBarrier = ps.isBarrier;
|
|
let Uses = ps.Uses;
|
|
let Defs = ps.Defs;
|
|
let isConvergent = ps.isConvergent;
|
|
|
|
// encoding
|
|
bits<7> sdst;
|
|
bits<16> simm16;
|
|
bits<32> imm;
|
|
}
|
|
|
|
class SOPK_Real32<bits<5> op, SOPK_Pseudo ps, string name = ps.Mnemonic> :
|
|
SOPK_Real <ps, name>,
|
|
Enc32 {
|
|
let Inst{15-0} = simm16;
|
|
let Inst{22-16} = !if(ps.has_sdst, sdst, ?);
|
|
let Inst{27-23} = op;
|
|
let Inst{31-28} = 0xb; //encoding
|
|
}
|
|
|
|
class SOPK_Real64<bits<5> op, SOPK_Pseudo ps> :
|
|
SOPK_Real<ps>,
|
|
Enc64 {
|
|
let Inst{15-0} = simm16;
|
|
let Inst{22-16} = !if(ps.has_sdst, sdst, ?);
|
|
let Inst{27-23} = op;
|
|
let Inst{31-28} = 0xb; //encoding
|
|
let Inst{63-32} = imm;
|
|
}
|
|
|
|
class SOPKInstTable <bit is_sopk, string cmpOp = ""> {
|
|
bit IsSOPK = is_sopk;
|
|
string BaseCmpOp = cmpOp;
|
|
}
|
|
|
|
class SOPK_32 <string opName, list<dag> pattern=[]> : SOPK_Pseudo <
|
|
opName,
|
|
(outs SReg_32:$sdst),
|
|
(ins s16imm:$simm16),
|
|
"$sdst, $simm16",
|
|
pattern>;
|
|
|
|
class SOPK_32_BR <string opName, list<dag> pattern=[]> : SOPK_Pseudo <
|
|
opName,
|
|
(outs),
|
|
(ins SOPPBrTarget:$simm16, SReg_32:$sdst),
|
|
"$sdst, $simm16",
|
|
pattern> {
|
|
let Defs = [EXEC];
|
|
let Uses = [EXEC];
|
|
let isBranch = 1;
|
|
let isTerminator = 1;
|
|
let SchedRW = [WriteBranch];
|
|
}
|
|
|
|
class SOPK_SCC <string opName, string base_op, bit isSignExt> : SOPK_Pseudo <
|
|
opName,
|
|
(outs),
|
|
!if(isSignExt,
|
|
(ins SReg_32:$sdst, s16imm:$simm16),
|
|
(ins SReg_32:$sdst, u16imm:$simm16)),
|
|
"$sdst, $simm16">,
|
|
SOPKInstTable<1, base_op>{
|
|
let Defs = [SCC];
|
|
}
|
|
|
|
class SOPK_32TIE <string opName, list<dag> pattern=[]> : SOPK_Pseudo <
|
|
opName,
|
|
(outs SReg_32:$sdst),
|
|
(ins SReg_32:$src0, s16imm:$simm16),
|
|
"$sdst, $simm16",
|
|
pattern
|
|
>;
|
|
|
|
let isReMaterializable = 1, isMoveImm = 1 in {
|
|
def S_MOVK_I32 : SOPK_32 <"s_movk_i32">;
|
|
} // End isReMaterializable = 1
|
|
let Uses = [SCC] in {
|
|
def S_CMOVK_I32 : SOPK_32 <"s_cmovk_i32">;
|
|
}
|
|
|
|
let isCompare = 1 in {
|
|
|
|
// This instruction is disabled for now until we can figure out how to teach
|
|
// the instruction selector to correctly use the S_CMP* vs V_CMP*
|
|
// instructions.
|
|
//
|
|
// When this instruction is enabled the code generator sometimes produces this
|
|
// invalid sequence:
|
|
//
|
|
// SCC = S_CMPK_EQ_I32 SGPR0, imm
|
|
// VCC = COPY SCC
|
|
// VGPR0 = V_CNDMASK VCC, VGPR0, VGPR1
|
|
//
|
|
// def S_CMPK_EQ_I32 : SOPK_SCC <"s_cmpk_eq_i32",
|
|
// [(set i1:$dst, (setcc i32:$src0, imm:$src1, SETEQ))]
|
|
// >;
|
|
|
|
def S_CMPK_EQ_I32 : SOPK_SCC <"s_cmpk_eq_i32", "s_cmp_eq_i32", 1>;
|
|
def S_CMPK_LG_I32 : SOPK_SCC <"s_cmpk_lg_i32", "s_cmp_lg_i32", 1>;
|
|
def S_CMPK_GT_I32 : SOPK_SCC <"s_cmpk_gt_i32", "s_cmp_gt_i32", 1>;
|
|
def S_CMPK_GE_I32 : SOPK_SCC <"s_cmpk_ge_i32", "s_cmp_ge_i32", 1>;
|
|
def S_CMPK_LT_I32 : SOPK_SCC <"s_cmpk_lt_i32", "s_cmp_lt_i32", 1>;
|
|
def S_CMPK_LE_I32 : SOPK_SCC <"s_cmpk_le_i32", "s_cmp_le_i32", 1>;
|
|
|
|
def S_CMPK_EQ_U32 : SOPK_SCC <"s_cmpk_eq_u32", "s_cmp_eq_u32", 0>;
|
|
def S_CMPK_LG_U32 : SOPK_SCC <"s_cmpk_lg_u32", "s_cmp_lg_u32", 0>;
|
|
def S_CMPK_GT_U32 : SOPK_SCC <"s_cmpk_gt_u32", "s_cmp_gt_u32", 0>;
|
|
def S_CMPK_GE_U32 : SOPK_SCC <"s_cmpk_ge_u32", "s_cmp_ge_u32", 0>;
|
|
def S_CMPK_LT_U32 : SOPK_SCC <"s_cmpk_lt_u32", "s_cmp_lt_u32", 0>;
|
|
def S_CMPK_LE_U32 : SOPK_SCC <"s_cmpk_le_u32", "s_cmp_le_u32", 0>;
|
|
} // End isCompare = 1
|
|
|
|
let isCommutable = 1, Constraints = "$sdst = $src0" in {
|
|
let Defs = [SCC] in
|
|
def S_ADDK_I32 : SOPK_32TIE <"s_addk_i32">;
|
|
def S_MULK_I32 : SOPK_32TIE <"s_mulk_i32">;
|
|
}
|
|
|
|
let SubtargetPredicate = isGFX6GFX7GFX8GFX9 in
|
|
def S_CBRANCH_I_FORK : SOPK_Pseudo <
|
|
"s_cbranch_i_fork",
|
|
(outs), (ins SReg_64:$sdst, SOPPBrTarget:$simm16),
|
|
"$sdst, $simm16"
|
|
>;
|
|
|
|
// FIXME: Need to truncate immediate to 16-bits.
|
|
class S_GETREG_B32_Pseudo<list<dag> pattern=[]> : SOPK_Pseudo <
|
|
"s_getreg_b32",
|
|
(outs SReg_32:$sdst), (ins hwreg:$simm16),
|
|
"$sdst, $simm16", pattern>;
|
|
|
|
// This is hasSideEffects to allow its use in readcyclecounter selection.
|
|
// FIXME: Should have separate pseudos for known may read MODE and
|
|
// only read MODE.
|
|
def S_GETREG_B32 : S_GETREG_B32_Pseudo<
|
|
[(set i32:$sdst, (int_amdgcn_s_getreg (i32 timm:$simm16)))]> {
|
|
let hasSideEffects = 1;
|
|
let Uses = [MODE];
|
|
}
|
|
|
|
// A version of the pseudo for reading hardware register fields that are
|
|
// known to remain the same during the course of the run. Has no side
|
|
// effects and doesn't read MODE.
|
|
def S_GETREG_B32_const : S_GETREG_B32_Pseudo;
|
|
|
|
let Defs = [MODE], Uses = [MODE] in {
|
|
|
|
// FIXME: Need to truncate immediate to 16-bits.
|
|
class S_SETREG_B32_Pseudo <list<dag> pattern=[]> : SOPK_Pseudo <
|
|
"s_setreg_b32",
|
|
(outs), (ins SReg_32:$sdst, hwreg:$simm16),
|
|
"$simm16, $sdst",
|
|
pattern>;
|
|
|
|
def S_SETREG_B32 : S_SETREG_B32_Pseudo <
|
|
[(int_amdgcn_s_setreg (i32 SIMM16bit:$simm16), i32:$sdst)]> {
|
|
// Use custom inserter to optimize some cases to
|
|
// S_DENORM_MODE/S_ROUND_MODE/S_SETREG_B32_mode.
|
|
let usesCustomInserter = 1;
|
|
let hasSideEffects = 1;
|
|
}
|
|
|
|
// Variant of SETREG that is guaranteed to only touch FP bits in the MODE
|
|
// register, so doesn't have unmodeled side effects.
|
|
def S_SETREG_B32_mode : S_SETREG_B32_Pseudo {
|
|
let hasSideEffects = 0;
|
|
}
|
|
|
|
// FIXME: Not on SI?
|
|
//def S_GETREG_REGRD_B32 : SOPK_32 <sopk<0x14, 0x13>, "s_getreg_regrd_b32">;
|
|
|
|
class S_SETREG_IMM32_B32_Pseudo : SOPK_Pseudo <
|
|
"s_setreg_imm32_b32",
|
|
(outs), (ins i32imm:$imm, hwreg:$simm16),
|
|
"$simm16, $imm"> {
|
|
let Size = 8; // Unlike every other SOPK instruction.
|
|
let has_sdst = 0;
|
|
}
|
|
|
|
def S_SETREG_IMM32_B32 : S_SETREG_IMM32_B32_Pseudo {
|
|
let hasSideEffects = 1;
|
|
}
|
|
|
|
// Variant of SETREG_IMM32 that is guaranteed to only touch FP bits in the MODE
|
|
// register, so doesn't have unmodeled side effects.
|
|
def S_SETREG_IMM32_B32_mode : S_SETREG_IMM32_B32_Pseudo {
|
|
let hasSideEffects = 0;
|
|
}
|
|
|
|
} // End Defs = [MODE], Uses = [MODE]
|
|
|
|
class SOPK_WAITCNT<string opName, list<dag> pat=[]> :
|
|
SOPK_Pseudo<
|
|
opName,
|
|
(outs),
|
|
(ins SReg_32:$sdst, s16imm:$simm16),
|
|
"$sdst, $simm16",
|
|
pat> {
|
|
let hasSideEffects = 1;
|
|
let mayLoad = 1;
|
|
let mayStore = 1;
|
|
let has_sdst = 1; // First source takes place of sdst in encoding
|
|
}
|
|
|
|
let SubtargetPredicate = isGFX9Plus in {
|
|
def S_CALL_B64 : SOPK_Pseudo<
|
|
"s_call_b64",
|
|
(outs SReg_64:$sdst),
|
|
(ins SOPPBrTarget:$simm16),
|
|
"$sdst, $simm16"> {
|
|
let isCall = 1;
|
|
}
|
|
} // End SubtargetPredicate = isGFX9Plus
|
|
|
|
def VersionImm : S16ImmOperand {
|
|
let DecoderMethod = "decodeVersionImm";
|
|
}
|
|
|
|
let SubtargetPredicate = isGFX10Plus in {
|
|
def S_VERSION : SOPK_Pseudo<
|
|
"s_version",
|
|
(outs),
|
|
(ins VersionImm:$simm16),
|
|
"$simm16"> {
|
|
let has_sdst = 0;
|
|
}
|
|
} // End SubtargetPredicate = isGFX10Plus
|
|
|
|
let SubtargetPredicate = isGFX10GFX11 in {
|
|
def S_SUBVECTOR_LOOP_BEGIN : SOPK_32_BR<"s_subvector_loop_begin">;
|
|
def S_SUBVECTOR_LOOP_END : SOPK_32_BR<"s_subvector_loop_end">;
|
|
|
|
def S_WAITCNT_VSCNT : SOPK_WAITCNT<"s_waitcnt_vscnt">;
|
|
def S_WAITCNT_VMCNT : SOPK_WAITCNT<"s_waitcnt_vmcnt">;
|
|
def S_WAITCNT_EXPCNT : SOPK_WAITCNT<"s_waitcnt_expcnt">;
|
|
def S_WAITCNT_LGKMCNT : SOPK_WAITCNT<"s_waitcnt_lgkmcnt">;
|
|
} // End SubtargetPredicate = isGFX10GFX11
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOPC Instructions
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
class SOPC_Pseudo<string opName, dag outs, dag ins,
|
|
string asmOps, list<dag> pattern=[]> :
|
|
SOP_Pseudo<opName, outs, ins, " " # asmOps, pattern> {
|
|
let mayLoad = 0;
|
|
let mayStore = 0;
|
|
let hasSideEffects = 0;
|
|
let SALU = 1;
|
|
let SOPC = 1;
|
|
let Defs = [SCC];
|
|
let SchedRW = [WriteSALU];
|
|
let UseNamedOperandTable = 1;
|
|
}
|
|
|
|
class SOPC_Real<bits<7> op, SOPC_Pseudo ps> :
|
|
InstSI <ps.OutOperandList, ps.InOperandList,
|
|
ps.Mnemonic # ps.AsmOperands>,
|
|
Enc32 {
|
|
let SALU = 1;
|
|
let SOPC = 1;
|
|
let isPseudo = 0;
|
|
let isCodeGenOnly = 0;
|
|
|
|
// copy relevant pseudo op flags
|
|
let SubtargetPredicate = ps.SubtargetPredicate;
|
|
let OtherPredicates = ps.OtherPredicates;
|
|
let AsmMatchConverter = ps.AsmMatchConverter;
|
|
let UseNamedOperandTable = ps.UseNamedOperandTable;
|
|
let TSFlags = ps.TSFlags;
|
|
let SchedRW = ps.SchedRW;
|
|
let mayLoad = ps.mayLoad;
|
|
let mayStore = ps.mayStore;
|
|
let Uses = ps.Uses;
|
|
let Defs = ps.Defs;
|
|
let isConvergent = ps.isConvergent;
|
|
|
|
// encoding
|
|
bits<8> src0;
|
|
bits<8> src1;
|
|
|
|
let Inst{7-0} = src0;
|
|
let Inst{15-8} = src1;
|
|
let Inst{22-16} = op;
|
|
let Inst{31-23} = 0x17e;
|
|
}
|
|
|
|
class SOPC_Base <RegisterOperand rc0, RegisterOperand rc1,
|
|
string opName, list<dag> pattern = []> : SOPC_Pseudo <
|
|
opName, (outs), (ins rc0:$src0, rc1:$src1),
|
|
"$src0, $src1", pattern > {
|
|
}
|
|
|
|
class SOPC_Helper <RegisterOperand rc, ValueType vt,
|
|
string opName, SDPatternOperator cond> : SOPC_Base <
|
|
rc, rc, opName,
|
|
[(set SCC, (UniformTernaryFrag<setcc> vt:$src0, vt:$src1, cond))] > {
|
|
}
|
|
|
|
class SOPC_CMP_32<string opName,
|
|
SDPatternOperator cond = COND_NULL, string revOp = opName>
|
|
: SOPC_Helper<SSrc_b32, i32, opName, cond>,
|
|
Commutable_REV<revOp, !eq(revOp, opName)>,
|
|
SOPKInstTable<0, opName> {
|
|
let isCompare = 1;
|
|
let isCommutable = 1;
|
|
}
|
|
|
|
class SOPC_CMP_F32<string opName,
|
|
SDPatternOperator cond = COND_NULL, string revOp = opName>
|
|
: SOPC_Helper<SSrc_b32, f32, opName, cond>,
|
|
Commutable_REV<revOp, !eq(revOp, opName)>,
|
|
SOPKInstTable<0, opName> {
|
|
let isCompare = 1;
|
|
let isCommutable = 1;
|
|
let mayRaiseFPException = 1;
|
|
let Uses = [MODE];
|
|
let SchedRW = [WriteSFPU];
|
|
}
|
|
|
|
class SOPC_CMP_F16<string opName,
|
|
SDPatternOperator cond = COND_NULL, string revOp = opName>
|
|
: SOPC_Helper<SSrc_b16, f16, opName, cond>,
|
|
Commutable_REV<revOp, !eq(revOp, opName)>,
|
|
SOPKInstTable<0, opName> {
|
|
let isCompare = 1;
|
|
let isCommutable = 1;
|
|
let mayRaiseFPException = 1;
|
|
let Uses = [MODE];
|
|
let SchedRW = [WriteSFPU];
|
|
}
|
|
|
|
class SOPC_CMP_64<string opName,
|
|
SDPatternOperator cond = COND_NULL, string revOp = opName>
|
|
: SOPC_Helper<SSrc_b64, i64, opName, cond>,
|
|
Commutable_REV<revOp, !eq(revOp, opName)> {
|
|
let isCompare = 1;
|
|
let isCommutable = 1;
|
|
}
|
|
|
|
class SOPC_32<string opName, list<dag> pattern = []>
|
|
: SOPC_Base<SSrc_b32, SSrc_b32, opName, pattern>;
|
|
|
|
class SOPC_64_32<string opName, list<dag> pattern = []>
|
|
: SOPC_Base<SSrc_b64, SSrc_b32, opName, pattern>;
|
|
|
|
def S_CMP_EQ_I32 : SOPC_CMP_32 <"s_cmp_eq_i32">;
|
|
def S_CMP_LG_I32 : SOPC_CMP_32 <"s_cmp_lg_i32">;
|
|
def S_CMP_GT_I32 : SOPC_CMP_32 <"s_cmp_gt_i32", COND_SGT>;
|
|
def S_CMP_GE_I32 : SOPC_CMP_32 <"s_cmp_ge_i32", COND_SGE>;
|
|
def S_CMP_LT_I32 : SOPC_CMP_32 <"s_cmp_lt_i32", COND_SLT, "s_cmp_gt_i32">;
|
|
def S_CMP_LE_I32 : SOPC_CMP_32 <"s_cmp_le_i32", COND_SLE, "s_cmp_ge_i32">;
|
|
def S_CMP_EQ_U32 : SOPC_CMP_32 <"s_cmp_eq_u32", COND_EQ>;
|
|
def S_CMP_LG_U32 : SOPC_CMP_32 <"s_cmp_lg_u32", COND_NE>;
|
|
def S_CMP_GT_U32 : SOPC_CMP_32 <"s_cmp_gt_u32", COND_UGT>;
|
|
def S_CMP_GE_U32 : SOPC_CMP_32 <"s_cmp_ge_u32", COND_UGE>;
|
|
def S_CMP_LT_U32 : SOPC_CMP_32 <"s_cmp_lt_u32", COND_ULT, "s_cmp_gt_u32">;
|
|
def S_CMP_LE_U32 : SOPC_CMP_32 <"s_cmp_le_u32", COND_ULE, "s_cmp_ge_u32">;
|
|
|
|
def S_BITCMP0_B32 : SOPC_32 <"s_bitcmp0_b32">;
|
|
def S_BITCMP1_B32 : SOPC_32 <"s_bitcmp1_b32">;
|
|
def S_BITCMP0_B64 : SOPC_64_32 <"s_bitcmp0_b64">;
|
|
def S_BITCMP1_B64 : SOPC_64_32 <"s_bitcmp1_b64">;
|
|
let SubtargetPredicate = isGFX6GFX7GFX8GFX9 in
|
|
def S_SETVSKIP : SOPC_32 <"s_setvskip">;
|
|
|
|
let SubtargetPredicate = isGFX8Plus in {
|
|
def S_CMP_EQ_U64 : SOPC_CMP_64 <"s_cmp_eq_u64", COND_EQ>;
|
|
def S_CMP_LG_U64 : SOPC_CMP_64 <"s_cmp_lg_u64", COND_NE>;
|
|
} // End SubtargetPredicate = isGFX8Plus
|
|
|
|
let SubtargetPredicate = HasVGPRIndexMode in {
|
|
// Setting the GPR index mode is really writing the fields in the mode
|
|
// register. We don't want to add mode register uses to every
|
|
// instruction, and it's too complicated to deal with anyway. This is
|
|
// modeled just as a side effect.
|
|
def S_SET_GPR_IDX_ON : SOPC_Pseudo <
|
|
"s_set_gpr_idx_on" ,
|
|
(outs),
|
|
(ins SSrc_b32:$src0, GPRIdxMode:$src1),
|
|
"$src0, $src1"> {
|
|
let Defs = [M0, MODE]; // No scc def
|
|
let Uses = [M0, MODE]; // Other bits of mode, m0 unmodified.
|
|
let hasSideEffects = 1; // Sets mode.gpr_idx_en
|
|
let FixedSize = 1;
|
|
}
|
|
}
|
|
|
|
let SubtargetPredicate = HasSALUFloatInsts in {
|
|
|
|
def S_CMP_LT_F32 : SOPC_CMP_F32<"s_cmp_lt_f32", COND_OLT, "s_cmp_gt_f32">;
|
|
def S_CMP_EQ_F32 : SOPC_CMP_F32<"s_cmp_eq_f32", COND_OEQ>;
|
|
def S_CMP_LE_F32 : SOPC_CMP_F32<"s_cmp_le_f32", COND_OLE, "s_cmp_ge_f32">;
|
|
def S_CMP_GT_F32 : SOPC_CMP_F32<"s_cmp_gt_f32", COND_OGT>;
|
|
def S_CMP_LG_F32 : SOPC_CMP_F32<"s_cmp_lg_f32", COND_ONE>;
|
|
def S_CMP_GE_F32 : SOPC_CMP_F32<"s_cmp_ge_f32", COND_OGE>;
|
|
def S_CMP_O_F32 : SOPC_CMP_F32<"s_cmp_o_f32", COND_O>;
|
|
def S_CMP_U_F32 : SOPC_CMP_F32<"s_cmp_u_f32", COND_UO>;
|
|
def S_CMP_NGE_F32 : SOPC_CMP_F32<"s_cmp_nge_f32", COND_ULT, "s_cmp_nle_f32">;
|
|
def S_CMP_NLG_F32 : SOPC_CMP_F32<"s_cmp_nlg_f32", COND_UEQ>;
|
|
def S_CMP_NGT_F32 : SOPC_CMP_F32<"s_cmp_ngt_f32", COND_ULE, "s_cmp_nlt_f32">;
|
|
def S_CMP_NLE_F32 : SOPC_CMP_F32<"s_cmp_nle_f32", COND_UGT>;
|
|
def S_CMP_NEQ_F32 : SOPC_CMP_F32<"s_cmp_neq_f32", COND_UNE>;
|
|
def S_CMP_NLT_F32 : SOPC_CMP_F32<"s_cmp_nlt_f32", COND_UGE>;
|
|
|
|
def S_CMP_LT_F16 : SOPC_CMP_F16<"s_cmp_lt_f16", COND_OLT, "s_cmp_gt_f16">;
|
|
def S_CMP_EQ_F16 : SOPC_CMP_F16<"s_cmp_eq_f16", COND_OEQ>;
|
|
def S_CMP_LE_F16 : SOPC_CMP_F16<"s_cmp_le_f16", COND_OLE, "s_cmp_ge_f16">;
|
|
def S_CMP_GT_F16 : SOPC_CMP_F16<"s_cmp_gt_f16", COND_OGT>;
|
|
def S_CMP_LG_F16 : SOPC_CMP_F16<"s_cmp_lg_f16", COND_ONE>;
|
|
def S_CMP_GE_F16 : SOPC_CMP_F16<"s_cmp_ge_f16", COND_OGE>;
|
|
def S_CMP_O_F16 : SOPC_CMP_F16<"s_cmp_o_f16", COND_O>;
|
|
def S_CMP_U_F16 : SOPC_CMP_F16<"s_cmp_u_f16", COND_UO>;
|
|
def S_CMP_NGE_F16 : SOPC_CMP_F16<"s_cmp_nge_f16", COND_ULT, "s_cmp_nle_f16">;
|
|
def S_CMP_NLG_F16 : SOPC_CMP_F16<"s_cmp_nlg_f16", COND_UEQ>;
|
|
def S_CMP_NGT_F16 : SOPC_CMP_F16<"s_cmp_ngt_f16", COND_ULE, "s_cmp_nlt_f16">;
|
|
def S_CMP_NLE_F16 : SOPC_CMP_F16<"s_cmp_nle_f16", COND_UGT>;
|
|
def S_CMP_NEQ_F16 : SOPC_CMP_F16<"s_cmp_neq_f16", COND_UNE>;
|
|
def S_CMP_NLT_F16 : SOPC_CMP_F16<"s_cmp_nlt_f16", COND_UGE>;
|
|
|
|
} // End SubtargetPredicate = HasSALUFloatInsts
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOPP Instructions
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
class SOPP_Pseudo<string opName, dag ins,
|
|
string asmOps = "", list<dag> pattern=[],
|
|
string sep = !if(!empty(asmOps), "", " "),
|
|
string keyName = opName> :
|
|
SOP_Pseudo<opName, (outs), ins, sep # asmOps, pattern> {
|
|
let mayLoad = 0;
|
|
let mayStore = 0;
|
|
let hasSideEffects = 0;
|
|
let SALU = 1;
|
|
let SOPP = 1;
|
|
let FixedSize = 1;
|
|
let SchedRW = [WriteSALU];
|
|
let UseNamedOperandTable = 1;
|
|
bits <16> simm16;
|
|
bits <1> fixed_imm = 0;
|
|
string KeyName = keyName;
|
|
}
|
|
|
|
class SOPPRelaxTable <bit isRelaxed, string keyName, string gfxip> {
|
|
bit IsRelaxed = isRelaxed;
|
|
string KeyName = keyName # gfxip;
|
|
}
|
|
|
|
class SOPP_Real<SOPP_Pseudo ps, string name = ps.Mnemonic> :
|
|
InstSI <ps.OutOperandList, ps.InOperandList,
|
|
name # ps.AsmOperands> {
|
|
let SALU = 1;
|
|
let SOPP = 1;
|
|
let isPseudo = 0;
|
|
let isCodeGenOnly = 0;
|
|
|
|
// copy relevant pseudo op flags
|
|
let SubtargetPredicate = ps.SubtargetPredicate;
|
|
let OtherPredicates = ps.OtherPredicates;
|
|
let AsmMatchConverter = ps.AsmMatchConverter;
|
|
let UseNamedOperandTable = ps.UseNamedOperandTable;
|
|
let TSFlags = ps.TSFlags;
|
|
let SchedRW = ps.SchedRW;
|
|
let mayLoad = ps.mayLoad;
|
|
let mayStore = ps.mayStore;
|
|
let isTerminator = ps.isTerminator;
|
|
let isReturn = ps.isReturn;
|
|
let isCall = ps.isCall;
|
|
let isBranch = ps.isBranch;
|
|
let isBarrier = ps.isBarrier;
|
|
let Uses = ps.Uses;
|
|
let Defs = ps.Defs;
|
|
let isConvergent = ps.isConvergent;
|
|
bits <16> simm16;
|
|
}
|
|
|
|
class SOPP_Real_32 <bits<7> op, SOPP_Pseudo ps, string name = ps.Mnemonic> : SOPP_Real<ps, name>,
|
|
Enc32 {
|
|
let Inst{15-0} = !if(ps.fixed_imm, ps.simm16, simm16);
|
|
let Inst{22-16} = op;
|
|
let Inst{31-23} = 0x17f;
|
|
}
|
|
|
|
class SOPP_Real_64 <bits<7> op, SOPP_Pseudo ps, string name = ps.Mnemonic> : SOPP_Real<ps, name>,
|
|
Enc64 {
|
|
// encoding
|
|
let Inst{15-0} = !if(ps.fixed_imm, ps.simm16, simm16);
|
|
let Inst{22-16} = op;
|
|
let Inst{31-23} = 0x17f;
|
|
//effectively a nop
|
|
let Inst{47-32} = 0x0;
|
|
let Inst{54-48} = 0x0;
|
|
let Inst{63-55} = 0x17f;
|
|
}
|
|
|
|
multiclass SOPP_With_Relaxation <string opName, dag ins,
|
|
string asmOps, list<dag> pattern=[]> {
|
|
def "" : SOPP_Pseudo <opName, ins, asmOps, pattern>;
|
|
def _pad_s_nop : SOPP_Pseudo <opName # "_pad_s_nop", ins, asmOps, pattern, " ", opName>;
|
|
}
|
|
|
|
def S_NOP : SOPP_Pseudo<"s_nop" , (ins i16imm:$simm16), "$simm16",
|
|
[(int_amdgcn_s_nop timm:$simm16)]> {
|
|
let hasSideEffects = 1;
|
|
}
|
|
|
|
let isTerminator = 1 in {
|
|
def S_ENDPGM : SOPP_Pseudo<"s_endpgm", (ins Endpgm:$simm16), "$simm16", [], ""> {
|
|
let isBarrier = 1;
|
|
let isReturn = 1;
|
|
let hasSideEffects = 1;
|
|
}
|
|
|
|
def S_ENDPGM_SAVED : SOPP_Pseudo<"s_endpgm_saved", (ins)> {
|
|
let SubtargetPredicate = isGFX8Plus;
|
|
let simm16 = 0;
|
|
let fixed_imm = 1;
|
|
let isBarrier = 1;
|
|
let isReturn = 1;
|
|
}
|
|
|
|
let SubtargetPredicate = isGFX9GFX10GFX11 in {
|
|
let isBarrier = 1, isReturn = 1, simm16 = 0, fixed_imm = 1 in {
|
|
def S_ENDPGM_ORDERED_PS_DONE :
|
|
SOPP_Pseudo<"s_endpgm_ordered_ps_done", (ins)>;
|
|
} // End isBarrier = 1, isReturn = 1, simm16 = 0, fixed_imm = 1
|
|
} // End SubtargetPredicate = isGFX9GFX10GFX11
|
|
|
|
let SubtargetPredicate = isGFX10Plus in {
|
|
let isBarrier = 1, isReturn = 1, simm16 = 0, fixed_imm = 1 in {
|
|
def S_CODE_END :
|
|
SOPP_Pseudo<"s_code_end", (ins)>;
|
|
} // End isBarrier = 1, isReturn = 1, simm16 = 0, fixed_imm = 1
|
|
} // End SubtargetPredicate = isGFX10Plus
|
|
|
|
let isBranch = 1, SchedRW = [WriteBranch] in {
|
|
let isBarrier = 1 in {
|
|
defm S_BRANCH : SOPP_With_Relaxation<
|
|
"s_branch" , (ins SOPPBrTarget:$simm16), "$simm16",
|
|
[(br bb:$simm16)]>;
|
|
}
|
|
|
|
let Uses = [SCC] in {
|
|
defm S_CBRANCH_SCC0 : SOPP_With_Relaxation<
|
|
"s_cbranch_scc0" , (ins SOPPBrTarget:$simm16),
|
|
"$simm16"
|
|
>;
|
|
defm S_CBRANCH_SCC1 : SOPP_With_Relaxation <
|
|
"s_cbranch_scc1" , (ins SOPPBrTarget:$simm16),
|
|
"$simm16"
|
|
>;
|
|
} // End Uses = [SCC]
|
|
|
|
let Uses = [VCC] in {
|
|
defm S_CBRANCH_VCCZ : SOPP_With_Relaxation <
|
|
"s_cbranch_vccz" , (ins SOPPBrTarget:$simm16),
|
|
"$simm16"
|
|
>;
|
|
defm S_CBRANCH_VCCNZ : SOPP_With_Relaxation <
|
|
"s_cbranch_vccnz" , (ins SOPPBrTarget:$simm16),
|
|
"$simm16"
|
|
>;
|
|
} // End Uses = [VCC]
|
|
|
|
let Uses = [EXEC] in {
|
|
defm S_CBRANCH_EXECZ : SOPP_With_Relaxation <
|
|
"s_cbranch_execz" , (ins SOPPBrTarget:$simm16),
|
|
"$simm16"
|
|
>;
|
|
defm S_CBRANCH_EXECNZ : SOPP_With_Relaxation <
|
|
"s_cbranch_execnz" , (ins SOPPBrTarget:$simm16),
|
|
"$simm16"
|
|
>;
|
|
} // End Uses = [EXEC]
|
|
|
|
defm S_CBRANCH_CDBGSYS : SOPP_With_Relaxation <
|
|
"s_cbranch_cdbgsys" , (ins SOPPBrTarget:$simm16),
|
|
"$simm16"
|
|
>;
|
|
|
|
defm S_CBRANCH_CDBGSYS_AND_USER : SOPP_With_Relaxation <
|
|
"s_cbranch_cdbgsys_and_user" , (ins SOPPBrTarget:$simm16),
|
|
"$simm16"
|
|
>;
|
|
|
|
defm S_CBRANCH_CDBGSYS_OR_USER : SOPP_With_Relaxation <
|
|
"s_cbranch_cdbgsys_or_user" , (ins SOPPBrTarget:$simm16),
|
|
"$simm16"
|
|
>;
|
|
|
|
defm S_CBRANCH_CDBGUSER : SOPP_With_Relaxation <
|
|
"s_cbranch_cdbguser" , (ins SOPPBrTarget:$simm16),
|
|
"$simm16"
|
|
>;
|
|
|
|
} // End isBranch = 1
|
|
} // End isTerminator = 1
|
|
|
|
let hasSideEffects = 1 in {
|
|
def S_BARRIER : SOPP_Pseudo <"s_barrier", (ins), "",
|
|
[(int_amdgcn_s_barrier)]> {
|
|
let SchedRW = [WriteBarrier];
|
|
let simm16 = 0;
|
|
let fixed_imm = 1;
|
|
let isConvergent = 1;
|
|
}
|
|
|
|
def S_BARRIER_WAIT : SOPP_Pseudo <"s_barrier_wait", (ins i16imm:$simm16), "$simm16",
|
|
[(int_amdgcn_s_barrier_wait timm:$simm16)]> {
|
|
let SchedRW = [WriteBarrier];
|
|
let isConvergent = 1;
|
|
}
|
|
|
|
|
|
let SchedRW = [WriteBarrier], isConvergent = 1 in {
|
|
let SubtargetPredicate = isGFX12Only in
|
|
def S_BARRIER_LEAVE : SOPP_Pseudo <"s_barrier_leave",
|
|
(ins), "", [(int_amdgcn_s_barrier_leave (i16 srcvalue))] > {
|
|
let simm16 = 0;
|
|
let fixed_imm = 1;
|
|
let Defs = [SCC];
|
|
}
|
|
|
|
let SubtargetPredicate = HasSBarrierLeaveImm in
|
|
def S_BARRIER_LEAVE_IMM : SOPP_Pseudo <"s_barrier_leave",
|
|
(ins i16imm:$simm16), "$simm16", [(int_amdgcn_s_barrier_leave timm:$simm16)]>;
|
|
}
|
|
|
|
def S_WAKEUP : SOPP_Pseudo <"s_wakeup", (ins) > {
|
|
let SubtargetPredicate = isGFX8GFX9GFX10GFX11GFX12;
|
|
let simm16 = 0;
|
|
let fixed_imm = 1;
|
|
let mayLoad = 1;
|
|
let mayStore = 1;
|
|
}
|
|
|
|
let SubtargetPredicate = HasSWakeupImm in {
|
|
def S_WAKEUP_imm : SOPP_Pseudo <"s_wakeup",
|
|
(ins i16imm:$simm16), "$simm16">;
|
|
} // End SubtargetPredicate = HasSWakeupImm
|
|
|
|
let SubtargetPredicate = isNotGFX1250Plus in {
|
|
def S_WAITCNT : SOPP_Pseudo <"s_waitcnt" , (ins SWaitCnt:$simm16), "$simm16",
|
|
[(int_amdgcn_s_waitcnt timm:$simm16)]>;
|
|
}
|
|
|
|
// "_soft" waitcnts are waitcnts that are either relaxed into their non-soft
|
|
// counterpart, or completely removed.
|
|
//
|
|
// These are inserted by SIMemoryLegalizer to resolve memory dependencies
|
|
// and later optimized by SIInsertWaitcnts
|
|
// For example, a S_WAITCNT_soft 0 can be completely removed in a function
|
|
// that doesn't access memory.
|
|
def S_WAITCNT_soft : SOPP_Pseudo <"s_soft_waitcnt" , (ins SWaitCnt:$simm16), "$simm16">;
|
|
def S_WAITCNT_VSCNT_soft : SOPK_WAITCNT<"s_soft_waitcnt_vscnt">;
|
|
let SubtargetPredicate = isGFX12Plus in {
|
|
def S_WAIT_LOADCNT_soft : SOPP_Pseudo <"s_soft_wait_loadcnt", (ins s16imm:$simm16), "$simm16">;
|
|
def S_WAIT_STORECNT_soft : SOPP_Pseudo <"s_soft_wait_storecnt", (ins s16imm:$simm16), "$simm16">;
|
|
let OtherPredicates = [HasImageInsts] in {
|
|
def S_WAIT_SAMPLECNT_soft : SOPP_Pseudo <"s_soft_wait_samplecnt", (ins s16imm:$simm16), "$simm16">;
|
|
def S_WAIT_BVHCNT_soft : SOPP_Pseudo <"s_soft_wait_bvhcnt", (ins s16imm:$simm16), "$simm16">;
|
|
} // End OtherPredicates = [HasImageInsts].
|
|
def S_WAIT_DSCNT_soft : SOPP_Pseudo <"s_soft_wait_dscnt", (ins s16imm:$simm16), "$simm16">;
|
|
def S_WAIT_KMCNT_soft : SOPP_Pseudo <"s_soft_wait_kmcnt", (ins s16imm:$simm16), "$simm16">;
|
|
}
|
|
|
|
|
|
let SubtargetPredicate = HasWaitXcnt in {
|
|
def S_WAIT_XCNT_soft : SOPP_Pseudo<"", (ins s16imm:$simm16), "$simm16">;
|
|
}
|
|
|
|
// Represents the point at which a wave must wait for all outstanding direct loads to LDS.
|
|
// Typically inserted by the memory legalizer and consumed by SIInsertWaitcnts.
|
|
def S_WAITCNT_lds_direct : SPseudoInstSI<(outs), (ins)> {
|
|
let hasSideEffects = 0;
|
|
}
|
|
|
|
let SubtargetPredicate = HasAsyncMark in {
|
|
def ASYNCMARK : SPseudoInstSI<(outs), (ins),
|
|
[(int_amdgcn_asyncmark)]> {
|
|
let maybeAtomic = 0;
|
|
}
|
|
def WAIT_ASYNCMARK : SOPP_Pseudo <"", (ins s16imm:$simm16), "$simm16",
|
|
[(int_amdgcn_wait_asyncmark timm:$simm16)]> {
|
|
let maybeAtomic = 0;
|
|
}
|
|
}
|
|
|
|
def S_SETHALT : SOPP_Pseudo <"s_sethalt" , (ins i32imm:$simm16), "$simm16",
|
|
[(int_amdgcn_s_sethalt timm:$simm16)]>;
|
|
def S_SETKILL : SOPP_Pseudo <"s_setkill" , (ins i16imm:$simm16), "$simm16"> {
|
|
let SubtargetPredicate = isNotGFX1250Plus;
|
|
}
|
|
|
|
// On SI the documentation says sleep for approximately 64 * low 2
|
|
// bits, consistent with the reported maximum of 448. On VI the
|
|
// maximum reported is 960 cycles, so 960 / 64 = 15 max, so is the
|
|
// maximum really 15 on VI?
|
|
def S_SLEEP : SOPP_Pseudo <"s_sleep", (ins i32imm:$simm16),
|
|
"$simm16", [(int_amdgcn_s_sleep timm:$simm16)]> {
|
|
}
|
|
|
|
def S_SLEEP_VAR : SOP1_0_32 <"s_sleep_var", [(int_amdgcn_s_sleep_var SSrc_b32:$src0)]> {
|
|
let hasSideEffects = 1;
|
|
}
|
|
|
|
def S_SETPRIO : SOPP_Pseudo <"s_setprio", (ins i16imm:$simm16), "$simm16",
|
|
[(int_amdgcn_s_setprio timm:$simm16)]> {
|
|
}
|
|
|
|
def S_SETPRIO_INC_WG : SOPP_Pseudo <"s_setprio_inc_wg", (ins i16imm:$simm16), "$simm16",
|
|
[(int_amdgcn_s_setprio_inc_wg timm:$simm16)]> {
|
|
let SubtargetPredicate = HasSetPrioIncWgInst;
|
|
}
|
|
|
|
def S_GET_SHADER_CYCLES_U64 : SOP1_64_0 <"s_get_shader_cycles_u64",
|
|
[(set i64:$sdst, (readcyclecounter))]> {
|
|
let SubtargetPredicate = HasSGetShaderCyclesInst;
|
|
let hasSideEffects = 1;
|
|
}
|
|
|
|
let Uses = [EXEC, M0] in {
|
|
def S_SENDMSG : SOPP_Pseudo <"s_sendmsg" , (ins SendMsg:$simm16), "$simm16",
|
|
[(int_amdgcn_s_sendmsg (i32 timm:$simm16), M0)]> {
|
|
}
|
|
|
|
def S_SENDMSGHALT : SOPP_Pseudo <"s_sendmsghalt" , (ins SendMsg:$simm16), "$simm16",
|
|
[(int_amdgcn_s_sendmsghalt (i32 timm:$simm16), M0)]> {
|
|
}
|
|
|
|
} // End Uses = [EXEC, M0]
|
|
|
|
def S_TRAP : SOPP_Pseudo <"s_trap" , (ins i16imm:$simm16), "$simm16"> {
|
|
let isTrap = 1;
|
|
}
|
|
|
|
def S_ICACHE_INV : SOPP_Pseudo <"s_icache_inv", (ins)> {
|
|
let simm16 = 0;
|
|
let fixed_imm = 1;
|
|
}
|
|
def S_INCPERFLEVEL : SOPP_Pseudo <"s_incperflevel", (ins i32imm:$simm16), "$simm16",
|
|
[(int_amdgcn_s_incperflevel timm:$simm16)]> {
|
|
}
|
|
def S_DECPERFLEVEL : SOPP_Pseudo <"s_decperflevel", (ins i32imm:$simm16), "$simm16",
|
|
[(int_amdgcn_s_decperflevel timm:$simm16)]> {
|
|
}
|
|
|
|
let Uses = [M0] in
|
|
def S_TTRACEDATA : SOPP_Pseudo <"s_ttracedata", (ins), "",
|
|
[(int_amdgcn_s_ttracedata M0)]> {
|
|
let simm16 = 0;
|
|
let fixed_imm = 1;
|
|
}
|
|
|
|
let SubtargetPredicate = HasVGPRIndexMode in {
|
|
def S_SET_GPR_IDX_OFF : SOPP_Pseudo<"s_set_gpr_idx_off", (ins) > {
|
|
let simm16 = 0;
|
|
let fixed_imm = 1;
|
|
let Defs = [MODE];
|
|
let Uses = [MODE];
|
|
}
|
|
}
|
|
|
|
def S_MONITOR_SLEEP : SOPP_Pseudo <"s_monitor_sleep", (ins i16imm:$simm16), "$simm16",
|
|
[(int_amdgcn_s_monitor_sleep timm:$simm16)]> {
|
|
let SubtargetPredicate = isGFX1250Plus;
|
|
}
|
|
|
|
} // End hasSideEffects
|
|
|
|
let SubtargetPredicate = HasVGPRIndexMode in {
|
|
def S_SET_GPR_IDX_MODE : SOPP_Pseudo<"s_set_gpr_idx_mode", (ins GPRIdxMode:$simm16),
|
|
"$simm16"> {
|
|
let Defs = [M0, MODE];
|
|
let Uses = [MODE];
|
|
}
|
|
}
|
|
|
|
let SubtargetPredicate = isGFX10Plus in {
|
|
def S_INST_PREFETCH :
|
|
SOPP_Pseudo<"s_inst_prefetch", (ins s16imm:$simm16), "$simm16">;
|
|
def S_CLAUSE :
|
|
SOPP_Pseudo<"s_clause", (ins s16imm:$simm16), "$simm16">;
|
|
def S_WAIT_IDLE :
|
|
SOPP_Pseudo <"s_wait_idle", (ins)> {
|
|
let simm16 = 0;
|
|
let fixed_imm = 1;
|
|
}
|
|
def S_WAITCNT_DEPCTR :
|
|
SOPP_Pseudo <"s_waitcnt_depctr" , (ins DepCtr:$simm16), "$simm16">;
|
|
|
|
let hasSideEffects = 0, Uses = [MODE], Defs = [MODE] in {
|
|
def S_ROUND_MODE :
|
|
SOPP_Pseudo<"s_round_mode", (ins s16imm:$simm16), "$simm16">;
|
|
def S_DENORM_MODE :
|
|
SOPP_Pseudo<"s_denorm_mode", (ins i32imm:$simm16), "$simm16",
|
|
[(SIdenorm_mode (i32 timm:$simm16))]>;
|
|
}
|
|
|
|
let hasSideEffects = 1 in
|
|
def S_TTRACEDATA_IMM :
|
|
SOPP_Pseudo<"s_ttracedata_imm", (ins s16imm:$simm16), "$simm16",
|
|
[(int_amdgcn_s_ttracedata_imm timm:$simm16)]>;
|
|
} // End SubtargetPredicate = isGFX10Plus
|
|
|
|
let SubtargetPredicate = isGFX11Plus in {
|
|
let OtherPredicates = [HasExportInsts] in
|
|
def S_WAIT_EVENT : SOPP_Pseudo<"s_wait_event", (ins WaitEvent:$simm16),
|
|
"$simm16", [(int_amdgcn_s_wait_event timm:$simm16)]> {
|
|
let hasSideEffects = 1;
|
|
}
|
|
def S_DELAY_ALU : SOPP_Pseudo<"s_delay_alu", (ins SDelayALU:$simm16),
|
|
"$simm16">;
|
|
} // End SubtargetPredicate = isGFX11Plus
|
|
|
|
let SubtargetPredicate = isGFX12Plus, hasSideEffects = 1 in {
|
|
def S_WAIT_LOADCNT :
|
|
SOPP_Pseudo<"s_wait_loadcnt", (ins s16imm:$simm16), "$simm16",
|
|
[(int_amdgcn_s_wait_loadcnt timm:$simm16)]>;
|
|
def S_WAIT_LOADCNT_DSCNT :
|
|
SOPP_Pseudo<"s_wait_loadcnt_dscnt", (ins s16imm:$simm16), "$simm16">;
|
|
def S_WAIT_STORECNT :
|
|
SOPP_Pseudo<"s_wait_storecnt", (ins s16imm:$simm16), "$simm16",
|
|
[(int_amdgcn_s_wait_storecnt timm:$simm16)]>;
|
|
def S_WAIT_STORECNT_DSCNT :
|
|
SOPP_Pseudo<"s_wait_storecnt_dscnt", (ins s16imm:$simm16), "$simm16">;
|
|
let OtherPredicates = [HasImageInsts] in {
|
|
def S_WAIT_SAMPLECNT :
|
|
SOPP_Pseudo<"s_wait_samplecnt", (ins s16imm:$simm16), "$simm16",
|
|
[(int_amdgcn_s_wait_samplecnt timm:$simm16)]>;
|
|
def S_WAIT_BVHCNT :
|
|
SOPP_Pseudo<"s_wait_bvhcnt", (ins s16imm:$simm16), "$simm16",
|
|
[(int_amdgcn_s_wait_bvhcnt timm:$simm16)]>;
|
|
} // End OtherPredicates = [HasImageInsts].
|
|
let OtherPredicates = [HasExportInsts] in
|
|
def S_WAIT_EXPCNT :
|
|
SOPP_Pseudo<"s_wait_expcnt", (ins s16imm:$simm16), "$simm16",
|
|
[(int_amdgcn_s_wait_expcnt timm:$simm16)]>;
|
|
def S_WAIT_DSCNT :
|
|
SOPP_Pseudo<"s_wait_dscnt", (ins s16imm:$simm16), "$simm16",
|
|
[(int_amdgcn_s_wait_dscnt timm:$simm16)]>;
|
|
def S_WAIT_KMCNT :
|
|
SOPP_Pseudo<"s_wait_kmcnt", (ins s16imm:$simm16), "$simm16",
|
|
[(int_amdgcn_s_wait_kmcnt timm:$simm16)]>;
|
|
} // End SubtargetPredicate = isGFX12Plus, hasSideEffects = 1
|
|
|
|
let SubtargetPredicate = isGFX1250Plus, hasSideEffects = 1 in {
|
|
def S_WAIT_ASYNCCNT :
|
|
SOPP_Pseudo<"s_wait_asynccnt", (ins s16imm:$simm16), "$simm16",
|
|
[(int_amdgcn_s_wait_asynccnt timm:$simm16)]> {
|
|
let mayLoad = 1;
|
|
let mayStore = 1;
|
|
let maybeAtomic = 0;
|
|
let Uses = [ASYNCcnt];
|
|
let Defs = [ASYNCcnt];
|
|
}
|
|
def S_WAIT_TENSORCNT :
|
|
SOPP_Pseudo<"s_wait_tensorcnt", (ins s16imm:$simm16), "$simm16",
|
|
[(int_amdgcn_s_wait_tensorcnt timm:$simm16)]> {
|
|
let mayLoad = 1;
|
|
let mayStore = 1;
|
|
let maybeAtomic = 0;
|
|
let Uses = [TENSORcnt];
|
|
let Defs = [TENSORcnt];
|
|
}
|
|
} // End SubtargetPredicate = isGFX1250Plus, hasSideEffects = 1
|
|
|
|
let SubtargetPredicate = HasWaitXcnt, hasSideEffects = 1 in {
|
|
def S_WAIT_XCNT :
|
|
SOPP_Pseudo<"s_wait_xcnt", (ins s16imm:$simm16), "$simm16">;
|
|
} // End SubtargetPredicate = hasWaitXcnt, hasSideEffects = 1
|
|
|
|
let SubtargetPredicate = Has1024AddressableVGPRs in {
|
|
def S_SET_VGPR_MSB : SOPP_Pseudo<"s_set_vgpr_msb" , (ins i16imm:$simm16), "$simm16"> {
|
|
let hasSideEffects = 1;
|
|
let Defs = [MODE];
|
|
}
|
|
}
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOP1 Patterns
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
def : GCNPat <
|
|
(AMDGPUendpgm),
|
|
(S_ENDPGM (i16 0))
|
|
>;
|
|
|
|
def : GCNPat <
|
|
(int_amdgcn_endpgm),
|
|
(S_ENDPGM (i16 0))
|
|
>;
|
|
|
|
def : GCNPat <
|
|
(i64 (UniformUnaryFrag<ctpop> i64:$src)),
|
|
(i64 (REG_SEQUENCE SReg_64,
|
|
(i32 (COPY_TO_REGCLASS (S_BCNT1_I32_B64 $src), SReg_32)), sub0,
|
|
(S_MOV_B32 (i32 0)), sub1))
|
|
>;
|
|
|
|
def : GCNPat <
|
|
(i32 (UniformBinFrag<smax> i32:$x, (i32 (ineg i32:$x)))),
|
|
(S_ABS_I32 SReg_32:$x)
|
|
>;
|
|
|
|
def : GCNPat <
|
|
(i16 imm:$imm),
|
|
(S_MOV_B32 imm:$imm)
|
|
>;
|
|
|
|
def : GCNPat <
|
|
(v2i32 (UniformBinFrag<and> v2i32:$x, v2i32:$y)),
|
|
(S_AND_B64 SReg_64:$x, SReg_64:$y)
|
|
>;
|
|
|
|
def : GCNPat <
|
|
(v2i32 (UniformBinFrag<or> v2i32:$x, v2i32:$y)),
|
|
(S_OR_B64 SReg_64:$x, SReg_64:$y)
|
|
>;
|
|
|
|
def : GCNPat <
|
|
(v2i32 (UniformBinFrag<xor> v2i32:$x, v2i32:$y)),
|
|
(S_XOR_B64 SReg_64:$x, SReg_64:$y)
|
|
>;
|
|
|
|
// Same as a 32-bit inreg
|
|
def : GCNPat<
|
|
(i32 (UniformUnaryFrag<sext> i16:$src)),
|
|
(S_SEXT_I32_I16 $src)
|
|
>;
|
|
|
|
let SubtargetPredicate = isGFX11Plus in
|
|
def : GCNPat <(int_amdgcn_s_wait_event_export_ready), (S_WAIT_EVENT (i16 2))>;
|
|
|
|
// The first 10 bits of the mode register are the core FP mode on all
|
|
// subtargets.
|
|
//
|
|
// The high bits include additional fields, intermixed with some
|
|
// non-floating point environment information. We extract the full
|
|
// register and clear non-relevant bits.
|
|
//
|
|
// EXCP_EN covers floating point exceptions, but also some other
|
|
// non-FP exceptions.
|
|
//
|
|
// Bits 12-18 cover the relevant exception mask on all subtargets.
|
|
//
|
|
// FIXME: Bit 18 is int_div0, should this be in the FP environment? I
|
|
// think the only source is v_rcp_iflag_i32.
|
|
//
|
|
// On GFX9+:
|
|
// Bit 23 is the additional FP16_OVFL mode.
|
|
//
|
|
// Bits 19, 20, and 21 cover non-FP exceptions and differ between
|
|
// gfx9/10/11, so we ignore them here.
|
|
|
|
// TODO: Would it be cheaper to emit multiple s_getreg_b32 calls for
|
|
// the ranges and combine the results?
|
|
|
|
defvar fp_round_mask = !add(!shl(1, 4), -1);
|
|
defvar fp_denorm_mask = !shl(!add(!shl(1, 4), -1), 4);
|
|
defvar dx10_clamp_mask = !shl(1, 8);
|
|
defvar ieee_mode_mask = !shl(1, 9);
|
|
|
|
// Covers fp_round, fp_denorm, dx10_clamp, and IEEE bit.
|
|
defvar fpmode_mask =
|
|
!or(fp_round_mask, fp_denorm_mask, dx10_clamp_mask, ieee_mode_mask);
|
|
|
|
defvar fp_excp_en_mask = !shl(!add(!shl(1, 7), -1), 12);
|
|
defvar fp16_ovfl = !shl(1, 23);
|
|
defvar fpmode_mask_gfx6plus = !or(fpmode_mask, fp_excp_en_mask);
|
|
defvar fpmode_mask_gfx9plus = !or(fpmode_mask_gfx6plus, fp16_ovfl);
|
|
|
|
class GetFPModePat<int fpmode_mask> : GCNPat<
|
|
(i32 get_fpmode),
|
|
(S_AND_B32 (i32 fpmode_mask),
|
|
(S_GETREG_B32 getHwRegImm<
|
|
HWREG.MODE, 0,
|
|
!add(!logtwo(fpmode_mask), 1)>.ret))
|
|
>;
|
|
|
|
// TODO: Might be worth moving to custom lowering so the and is
|
|
// exposed to demanded bits optimizations. Most users probably only
|
|
// care about the rounding or denorm mode bits. We also can reduce the
|
|
// demanded read from the getreg immediate.
|
|
let SubtargetPredicate = isGFX9Plus in {
|
|
// Last bit = FP16_OVFL
|
|
def : GetFPModePat<fpmode_mask_gfx9plus>;
|
|
}
|
|
|
|
// Last bit = EXCP_EN.int_div0
|
|
let SubtargetPredicate = isNotGFX9Plus in {
|
|
def : GetFPModePat<fpmode_mask_gfx6plus>;
|
|
}
|
|
|
|
let SubtargetPredicate = isGFX9GFX10 in
|
|
def : GCNPat<
|
|
(int_amdgcn_pops_exiting_wave_id),
|
|
(S_MOV_B32_sideeffects (i32 SRC_POPS_EXITING_WAVE_ID))
|
|
>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOP2 Patterns
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
def UniformSelect : PatFrag<
|
|
(ops node:$src0, node:$src1),
|
|
(select SCC, $src0, $src1),
|
|
[{ return !N->isDivergent(); }]
|
|
>;
|
|
|
|
let AddedComplexity = 20 in {
|
|
def : GCNPat<
|
|
(i32 (UniformSelect i32:$src0, i32:$src1)),
|
|
(S_CSELECT_B32 SSrc_b32:$src0, SSrc_b32:$src1)
|
|
>;
|
|
|
|
// TODO: The predicate should not be necessary, but enabling this pattern for
|
|
// all subtargets generates worse code in some cases.
|
|
let OtherPredicates = [HasPseudoScalarTrans] in
|
|
def : GCNPat<
|
|
(f32 (UniformSelect f32:$src0, f32:$src1)),
|
|
(S_CSELECT_B32 SSrc_b32:$src0, SSrc_b32:$src1)
|
|
>;
|
|
}
|
|
|
|
// V_ADD_I32_e32/S_ADD_U32 produces carry in VCC/SCC. For the vector
|
|
// case, the sgpr-copies pass will fix this to use the vector version.
|
|
def : GCNPat <
|
|
(i32 (addc i32:$src0, i32:$src1)),
|
|
(S_ADD_U32 $src0, $src1)
|
|
>;
|
|
|
|
// FIXME: We need to use COPY_TO_REGCLASS to work-around the fact that
|
|
// REG_SEQUENCE patterns don't support instructions with multiple
|
|
// outputs.
|
|
def : GCNPat<
|
|
(i64 (UniformUnaryFrag<zext> i16:$src)),
|
|
(REG_SEQUENCE SReg_64,
|
|
(i32 (COPY_TO_REGCLASS (S_AND_B32 $src, (S_MOV_B32 (i32 0xffff))), SGPR_32)), sub0,
|
|
(S_MOV_B32 (i32 0)), sub1)
|
|
>;
|
|
|
|
def : GCNPat <
|
|
(i64 (UniformUnaryFrag<sext> i16:$src)),
|
|
(REG_SEQUENCE SReg_64, (i32 (S_SEXT_I32_I16 $src)), sub0,
|
|
(i32 (COPY_TO_REGCLASS (S_ASHR_I32 (i32 (S_SEXT_I32_I16 $src)), (S_MOV_B32 (i32 31))), SGPR_32)), sub1)
|
|
>;
|
|
|
|
def : GCNPat<
|
|
(i32 (UniformUnaryFrag<zext> i16:$src)),
|
|
(S_AND_B32 (S_MOV_B32 (i32 0xffff)), $src)
|
|
>;
|
|
|
|
class ScalarNot2Pat<Instruction inst, SDPatternOperator op, ValueType vt,
|
|
SDPatternOperator notnode = !if(vt.isVector, vnot, not)> : GCNPat<
|
|
(UniformBinFrag<op> vt:$src0, (notnode vt:$src1)),
|
|
(inst getSOPSrcForVT<vt>.ret:$src0, getSOPSrcForVT<vt>.ret:$src1)
|
|
>;
|
|
|
|
// Match these for some more types
|
|
// TODO: i1
|
|
def : ScalarNot2Pat<S_ANDN2_B32, and, i16>;
|
|
def : ScalarNot2Pat<S_ANDN2_B32, and, v2i16>;
|
|
def : ScalarNot2Pat<S_ANDN2_B64, and, v4i16>;
|
|
def : ScalarNot2Pat<S_ANDN2_B64, and, v2i32>;
|
|
|
|
def : ScalarNot2Pat<S_ORN2_B32, or, i16>;
|
|
def : ScalarNot2Pat<S_ORN2_B32, or, v2i16>;
|
|
def : ScalarNot2Pat<S_ORN2_B64, or, v4i16>;
|
|
def : ScalarNot2Pat<S_ORN2_B64, or, v2i32>;
|
|
|
|
let WaveSizePredicate = isWave32 in {
|
|
def : GCNPat<
|
|
(i1 (not (or_oneuse i1:$src0, i1:$src1))),
|
|
(S_NOR_B32 i1:$src0, i1:$src1)
|
|
>;
|
|
}
|
|
|
|
let WaveSizePredicate = isWave64 in {
|
|
def : GCNPat<
|
|
(i1 (not (or_oneuse i1:$src0, i1:$src1))),
|
|
(S_NOR_B64 i1:$src0, i1:$src1)
|
|
>;
|
|
}
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// Target-specific instruction encodings.
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
class Select<GFXGen Gen, string opName> : SIMCInstr<opName, Gen.Subtarget> {
|
|
Predicate AssemblerPredicate = Gen.AssemblerPredicate;
|
|
string DecoderNamespace = Gen.DecoderNamespace;
|
|
}
|
|
|
|
class Select_vi<string opName> : SIMCInstr<opName, SIEncodingFamily.VI> {
|
|
Predicate AssemblerPredicate = isGFX8GFX9;
|
|
string DecoderNamespace = "GFX8";
|
|
}
|
|
|
|
class Select_gfx6_gfx7<string opName> : SIMCInstr<opName, SIEncodingFamily.SI> {
|
|
Predicate AssemblerPredicate = isGFX6GFX7;
|
|
string DecoderNamespace = "GFX6GFX7";
|
|
}
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOP1 - GFX13
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOP1_Real_gfx13<bits<8> op, string name = !tolower(NAME)> {
|
|
defvar ps = !cast<SOP1_Pseudo>(NAME);
|
|
def _gfx13 : SOP1_Real<op, ps, name>,
|
|
Select<GFX13Gen, ps.Mnemonic>;
|
|
if !ne(ps.Mnemonic, name) then
|
|
def : MnemonicAlias<ps.Mnemonic, name>, Requires<[isGFX13Only]>;
|
|
}
|
|
|
|
multiclass SOP1_M0_Real_gfx13<bits<8> op> {
|
|
defvar ps = !cast<SOP1_Pseudo>(NAME);
|
|
def _gfx13 : SOP1_Real<op, ps>, Select<GFX13Gen, ps.PseudoInstr> {
|
|
let Inst{7-0} = M0_gfx11plus.HWEncoding{7-0}; // Set Src0 encoding to M0
|
|
}
|
|
}
|
|
|
|
multiclass SOP1_IMM_Real_gfx13<bits<8> op> {
|
|
defvar ps = !cast<SOP1_Pseudo>(NAME);
|
|
def _gfx13 : SOP1_Real<op, ps>,
|
|
Select<GFX13Gen, ps.PseudoInstr>;
|
|
}
|
|
|
|
defm S_GET_SHADER_CYCLES_U64 : SOP1_Real_gfx13<0x011>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOP1 - GFX11, GFX12, GFX13
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOP1_Real_gfx11<bits<8> op, string name = !tolower(NAME)> {
|
|
defvar ps = !cast<SOP1_Pseudo>(NAME);
|
|
def _gfx11 : SOP1_Real<op, ps, name>,
|
|
Select<GFX11Gen, ps.PseudoInstr>;
|
|
if !ne(ps.Mnemonic, name) then
|
|
def : AMDGPUMnemonicAlias<ps.Mnemonic, name> {
|
|
let AssemblerPredicate = isGFX11Only;
|
|
}
|
|
}
|
|
|
|
multiclass SOP1_Real_gfx12<bits<8> op, string name = !tolower(NAME)> {
|
|
defvar ps = !cast<SOP1_Pseudo>(NAME);
|
|
def _gfx12 : SOP1_Real<op, ps, name>,
|
|
Select<GFX12Gen, ps.PseudoInstr>;
|
|
if !ne(ps.Mnemonic, name) then
|
|
def : AMDGPUMnemonicAlias<ps.Mnemonic, name> {
|
|
let AssemblerPredicate = isGFX12Only;
|
|
}
|
|
}
|
|
|
|
multiclass SOP1_M0_Real_gfx12<bits<8> op> {
|
|
defvar ps = !cast<SOP1_Pseudo>(NAME);
|
|
def _gfx12 : SOP1_Real<op, ps>, Select<GFX12Gen, ps.PseudoInstr> {
|
|
let Inst{7-0} = M0_gfx11plus.HWEncoding{7-0}; // Set Src0 encoding to M0
|
|
}
|
|
}
|
|
|
|
multiclass SOP1_M0_Real_gfx12_gfx13<bits<8> op> :
|
|
SOP1_M0_Real_gfx12<op>, SOP1_M0_Real_gfx13<op>;
|
|
|
|
multiclass SOP1_IMM_Real_gfx12<bits<8> op> {
|
|
defvar ps = !cast<SOP1_Pseudo>(NAME);
|
|
def _gfx12 : SOP1_Real<op, ps>,
|
|
Select<GFX12Gen, ps.PseudoInstr>;
|
|
}
|
|
|
|
multiclass SOP1_IMM_Real_gfx12_gfx13<bits<8> op> :
|
|
SOP1_IMM_Real_gfx12<op>, SOP1_IMM_Real_gfx13<op>;
|
|
|
|
multiclass SOP1_Real_gfx11_gfx12<bits<8> op, string name = !tolower(NAME)> :
|
|
SOP1_Real_gfx11<op, name>, SOP1_Real_gfx12<op, name>;
|
|
|
|
multiclass SOP1_Real_gfx1250<bits<8> op, string name = !tolower(NAME)> {
|
|
defvar ps = !cast<SOP1_Pseudo>(NAME);
|
|
def _gfx1250 : SOP1_Real<op, ps, name>,
|
|
Select<GFX1250Gen, ps.PseudoInstr>;
|
|
if !ne(ps.Mnemonic, name) then
|
|
let AssemblerPredicate = isGFX1250Plus in
|
|
def : AMDGPUMnemonicAlias<ps.Mnemonic, name>;
|
|
}
|
|
|
|
multiclass SOP1_Real_gfx11_gfx12_gfx13<bits<8> op> :
|
|
SOP1_Real_gfx11<op>, SOP1_Real_gfx12<op>, SOP1_Real_gfx13<op>;
|
|
|
|
multiclass SOP1_Real_gfx12_gfx13<bits<8> op> :
|
|
SOP1_Real_gfx12<op>, SOP1_Real_gfx13<op>;
|
|
|
|
defm S_MOV_B32 : SOP1_Real_gfx11_gfx12<0x000>;
|
|
defm S_MOV_B64 : SOP1_Real_gfx11_gfx12<0x001>;
|
|
defm S_CMOV_B32 : SOP1_Real_gfx11_gfx12<0x002>;
|
|
defm S_CMOV_B64 : SOP1_Real_gfx11_gfx12<0x003>;
|
|
defm S_BREV_B32 : SOP1_Real_gfx11_gfx12<0x004>;
|
|
defm S_BREV_B64 : SOP1_Real_gfx11_gfx12<0x005>;
|
|
defm S_FF1_I32_B32 : SOP1_Real_gfx11_gfx12<0x008, "s_ctz_i32_b32">;
|
|
defm S_FF1_I32_B64 : SOP1_Real_gfx11_gfx12<0x009, "s_ctz_i32_b64">;
|
|
defm S_FLBIT_I32_B32 : SOP1_Real_gfx11_gfx12<0x00a, "s_clz_i32_u32">;
|
|
defm S_FLBIT_I32_B64 : SOP1_Real_gfx11_gfx12<0x00b, "s_clz_i32_u64">;
|
|
defm S_FLBIT_I32 : SOP1_Real_gfx11_gfx12<0x00c, "s_cls_i32">;
|
|
defm S_FLBIT_I32_I64 : SOP1_Real_gfx11_gfx12<0x00d, "s_cls_i32_i64">;
|
|
defm S_SEXT_I32_I8 : SOP1_Real_gfx11_gfx12<0x00e>;
|
|
defm S_SEXT_I32_I16 : SOP1_Real_gfx11_gfx12<0x00f>;
|
|
defm S_BITSET0_B32 : SOP1_Real_gfx11_gfx12<0x010>;
|
|
defm S_BITSET0_B64 : SOP1_Real_gfx11_gfx12<0x011>;
|
|
defm S_BITSET1_B32 : SOP1_Real_gfx11_gfx12<0x012>;
|
|
defm S_BITSET1_B64 : SOP1_Real_gfx11_gfx12<0x013>;
|
|
defm S_BITREPLICATE_B64_B32 : SOP1_Real_gfx11_gfx12<0x014>;
|
|
defm S_ABS_I32 : SOP1_Real_gfx11_gfx12<0x015>;
|
|
defm S_BCNT0_I32_B32 : SOP1_Real_gfx11_gfx12<0x016>;
|
|
defm S_BCNT0_I32_B64 : SOP1_Real_gfx11_gfx12<0x017>;
|
|
defm S_BCNT1_I32_B32 : SOP1_Real_gfx11_gfx12<0x018>;
|
|
defm S_BCNT1_I32_B64 : SOP1_Real_gfx11_gfx12<0x019>;
|
|
defm S_QUADMASK_B32 : SOP1_Real_gfx11_gfx12<0x01a>;
|
|
defm S_QUADMASK_B64 : SOP1_Real_gfx11_gfx12<0x01b>;
|
|
defm S_WQM_B32 : SOP1_Real_gfx11_gfx12<0x01c>;
|
|
defm S_WQM_B64 : SOP1_Real_gfx11_gfx12<0x01d>;
|
|
defm S_NOT_B32 : SOP1_Real_gfx11_gfx12<0x01e>;
|
|
defm S_NOT_B64 : SOP1_Real_gfx11_gfx12<0x01f>;
|
|
defm S_AND_SAVEEXEC_B32 : SOP1_Real_gfx11_gfx12<0x020>;
|
|
defm S_AND_SAVEEXEC_B64 : SOP1_Real_gfx11_gfx12<0x021>;
|
|
defm S_OR_SAVEEXEC_B32 : SOP1_Real_gfx11_gfx12<0x022>;
|
|
defm S_OR_SAVEEXEC_B64 : SOP1_Real_gfx11_gfx12<0x023>;
|
|
defm S_XOR_SAVEEXEC_B32 : SOP1_Real_gfx11_gfx12<0x024>;
|
|
defm S_XOR_SAVEEXEC_B64 : SOP1_Real_gfx11_gfx12<0x025>;
|
|
defm S_NAND_SAVEEXEC_B32 : SOP1_Real_gfx11_gfx12<0x026>;
|
|
defm S_NAND_SAVEEXEC_B64 : SOP1_Real_gfx11_gfx12<0x027>;
|
|
defm S_NOR_SAVEEXEC_B32 : SOP1_Real_gfx11_gfx12<0x028>;
|
|
defm S_NOR_SAVEEXEC_B64 : SOP1_Real_gfx11_gfx12<0x029>;
|
|
defm S_XNOR_SAVEEXEC_B32 : SOP1_Real_gfx11_gfx12<0x02a>;
|
|
defm S_ANDN1_SAVEEXEC_B32 : SOP1_Real_gfx11_gfx12<0x02c, "s_and_not0_saveexec_b32">;
|
|
defm S_ANDN1_SAVEEXEC_B64 : SOP1_Real_gfx11_gfx12<0x02d, "s_and_not0_saveexec_b64">;
|
|
defm S_ORN1_SAVEEXEC_B32 : SOP1_Real_gfx11_gfx12<0x02e, "s_or_not0_saveexec_b32">;
|
|
defm S_ORN1_SAVEEXEC_B64 : SOP1_Real_gfx11_gfx12<0x02f, "s_or_not0_saveexec_b64">;
|
|
defm S_ANDN2_SAVEEXEC_B32 : SOP1_Real_gfx11_gfx12<0x030, "s_and_not1_saveexec_b32">;
|
|
defm S_ANDN2_SAVEEXEC_B64 : SOP1_Real_gfx11_gfx12<0x031, "s_and_not1_saveexec_b64">;
|
|
defm S_ORN2_SAVEEXEC_B32 : SOP1_Real_gfx11_gfx12<0x032, "s_or_not1_saveexec_b32">;
|
|
defm S_ORN2_SAVEEXEC_B64 : SOP1_Real_gfx11_gfx12<0x033, "s_or_not1_saveexec_b64">;
|
|
defm S_ANDN1_WREXEC_B32 : SOP1_Real_gfx11_gfx12<0x034, "s_and_not0_wrexec_b32">;
|
|
defm S_ANDN1_WREXEC_B64 : SOP1_Real_gfx11_gfx12<0x035, "s_and_not0_wrexec_b64">;
|
|
defm S_ANDN2_WREXEC_B32 : SOP1_Real_gfx11_gfx12<0x036, "s_and_not1_wrexec_b32">;
|
|
defm S_ANDN2_WREXEC_B64 : SOP1_Real_gfx11_gfx12<0x037, "s_and_not1_wrexec_b64">;
|
|
defm S_MOVRELS_B32 : SOP1_Real_gfx11_gfx12<0x040>;
|
|
defm S_MOVRELS_B64 : SOP1_Real_gfx11_gfx12<0x041>;
|
|
defm S_MOVRELD_B32 : SOP1_Real_gfx11_gfx12<0x042>;
|
|
defm S_MOVRELD_B64 : SOP1_Real_gfx11_gfx12<0x043>;
|
|
defm S_MOVRELSD_2_B32 : SOP1_Real_gfx11_gfx12<0x044>;
|
|
let OtherPredicates = [isNotGFX1250Plus] in {
|
|
defm S_GETPC_B64 : SOP1_Real_gfx11_gfx12<0x047>;
|
|
defm S_SETPC_B64 : SOP1_Real_gfx11_gfx12<0x048>;
|
|
defm S_SWAPPC_B64 : SOP1_Real_gfx11_gfx12<0x049>;
|
|
defm S_RFE_B64 : SOP1_Real_gfx11_gfx12<0x04a>;
|
|
}
|
|
defm S_GETPC_B64 : SOP1_Real_gfx1250<0x047, "s_get_pc_i64">;
|
|
defm S_SETPC_B64 : SOP1_Real_gfx1250<0x048, "s_set_pc_i64">;
|
|
defm S_SWAPPC_B64 : SOP1_Real_gfx1250<0x049, "s_swap_pc_i64">;
|
|
defm S_RFE_B64 : SOP1_Real_gfx1250<0x04a, "s_rfe_i64">;
|
|
defm S_SENDMSG_RTN_B32 : SOP1_Real_gfx11_gfx12_gfx13<0x04c>;
|
|
defm S_SENDMSG_RTN_B64 : SOP1_Real_gfx11_gfx12_gfx13<0x04d>;
|
|
defm S_BARRIER_SIGNAL_M0 : SOP1_M0_Real_gfx12_gfx13<0x04e>;
|
|
defm S_BARRIER_SIGNAL_ISFIRST_M0 : SOP1_M0_Real_gfx12_gfx13<0x04f>;
|
|
defm S_GET_BARRIER_STATE_M0 : SOP1_M0_Real_gfx12_gfx13<0x050>;
|
|
defm S_BARRIER_INIT_M0 : SOP1_M0_Real_gfx12_gfx13<0x051>;
|
|
defm S_BARRIER_JOIN_M0 : SOP1_M0_Real_gfx12_gfx13<0x052>;
|
|
defm S_BARRIER_SIGNAL_IMM : SOP1_IMM_Real_gfx12_gfx13<0x04e>;
|
|
defm S_BARRIER_SIGNAL_ISFIRST_IMM : SOP1_IMM_Real_gfx12_gfx13<0x04f>;
|
|
defm S_GET_BARRIER_STATE_IMM : SOP1_IMM_Real_gfx12_gfx13<0x050>;
|
|
defm S_BARRIER_INIT_IMM : SOP1_IMM_Real_gfx12_gfx13<0x051>;
|
|
defm S_BARRIER_JOIN_IMM : SOP1_IMM_Real_gfx12_gfx13<0x052>;
|
|
defm S_ALLOC_VGPR : SOP1_Real_gfx12_gfx13<0x053>;
|
|
defm S_SLEEP_VAR : SOP1_IMM_Real_gfx12_gfx13<0x058>;
|
|
|
|
// GFX1250, GFX13
|
|
defm S_GET_SHADER_CYCLES_U64 : SOP1_Real_gfx12<0x06>;
|
|
defm S_ADD_PC_I64 : SOP1_Real_gfx12_gfx13<0x04b>;
|
|
defm S_WAKEUP_BARRIER_M0 : SOP1_M0_Real_gfx12_gfx13<0x057>;
|
|
defm S_WAKEUP_BARRIER_IMM : SOP1_IMM_Real_gfx12_gfx13<0x057>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOP1 - GFX1150, GFX12, GFX13
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
defm S_CEIL_F32 : SOP1_Real_gfx11_gfx12_gfx13<0x060>;
|
|
defm S_FLOOR_F32 : SOP1_Real_gfx11_gfx12_gfx13<0x061>;
|
|
defm S_TRUNC_F32 : SOP1_Real_gfx11_gfx12_gfx13<0x062>;
|
|
defm S_RNDNE_F32 : SOP1_Real_gfx11_gfx12_gfx13<0x063>;
|
|
defm S_CVT_F32_I32 : SOP1_Real_gfx11_gfx12_gfx13<0x064>;
|
|
defm S_CVT_F32_U32 : SOP1_Real_gfx11_gfx12_gfx13<0x065>;
|
|
defm S_CVT_I32_F32 : SOP1_Real_gfx11_gfx12_gfx13<0x066>;
|
|
defm S_CVT_U32_F32 : SOP1_Real_gfx11_gfx12_gfx13<0x067>;
|
|
defm S_CVT_F16_F32 : SOP1_Real_gfx11_gfx12_gfx13<0x068>;
|
|
defm S_CVT_F32_F16 : SOP1_Real_gfx11_gfx12_gfx13<0x069>;
|
|
defm S_CVT_HI_F32_F16 : SOP1_Real_gfx11_gfx12_gfx13<0x06a>;
|
|
defm S_CEIL_F16 : SOP1_Real_gfx11_gfx12_gfx13<0x06b>;
|
|
defm S_FLOOR_F16 : SOP1_Real_gfx11_gfx12_gfx13<0x06c>;
|
|
defm S_TRUNC_F16 : SOP1_Real_gfx11_gfx12_gfx13<0x06d>;
|
|
defm S_RNDNE_F16 : SOP1_Real_gfx11_gfx12_gfx13<0x06e>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOP1 - GFX10, GFX13
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOP1_Real_gfx10<bits<8> op> {
|
|
defvar ps = !cast<SOP1_Pseudo>(NAME);
|
|
def _gfx10 : SOP1_Real<op, ps>,
|
|
Select<GFX10Gen, ps.PseudoInstr>;
|
|
}
|
|
|
|
multiclass SOP1_Real_gfx10_gfx13<bits<8> op> :
|
|
SOP1_Real_gfx10<op>, SOP1_Real_gfx13<op>;
|
|
|
|
multiclass SOP1_Real_gfx10_Renamed_gfx13<bits<8> op, string gfx13_name> :
|
|
SOP1_Real_gfx10<op>, SOP1_Real_gfx13<op, gfx13_name>;
|
|
|
|
defm S_ANDN1_SAVEEXEC_B64 : SOP1_Real_gfx10_Renamed_gfx13<0x37, "s_and_not0_saveexec_b64">;
|
|
defm S_ORN1_SAVEEXEC_B64 : SOP1_Real_gfx10_Renamed_gfx13<0x38, "s_or_not0_saveexec_b64">;
|
|
defm S_ANDN1_WREXEC_B64 : SOP1_Real_gfx10_Renamed_gfx13<0x39, "s_and_not0_wrexec_b64">;
|
|
defm S_ANDN2_WREXEC_B64 : SOP1_Real_gfx10_Renamed_gfx13<0x3a, "s_and_not1_wrexec_b64">;
|
|
defm S_BITREPLICATE_B64_B32 : SOP1_Real_gfx10_gfx13<0x03b>;
|
|
defm S_AND_SAVEEXEC_B32 : SOP1_Real_gfx10_gfx13<0x03c>;
|
|
defm S_OR_SAVEEXEC_B32 : SOP1_Real_gfx10_gfx13<0x03d>;
|
|
defm S_XOR_SAVEEXEC_B32 : SOP1_Real_gfx10_gfx13<0x03e>;
|
|
defm S_ANDN2_SAVEEXEC_B32 : SOP1_Real_gfx10_Renamed_gfx13<0x3f, "s_and_not1_saveexec_b32">;
|
|
defm S_ORN2_SAVEEXEC_B32 : SOP1_Real_gfx10_Renamed_gfx13<0x40, "s_or_not1_saveexec_b32">;
|
|
defm S_NAND_SAVEEXEC_B32 : SOP1_Real_gfx10_gfx13<0x041>;
|
|
defm S_NOR_SAVEEXEC_B32 : SOP1_Real_gfx10_gfx13<0x042>;
|
|
defm S_XNOR_SAVEEXEC_B32 : SOP1_Real_gfx10_gfx13<0x043>;
|
|
defm S_ANDN1_SAVEEXEC_B32 : SOP1_Real_gfx10_Renamed_gfx13<0x44, "s_and_not0_saveexec_b32">;
|
|
defm S_ORN1_SAVEEXEC_B32 : SOP1_Real_gfx10_Renamed_gfx13<0x45, "s_or_not0_saveexec_b32">;
|
|
defm S_ANDN1_WREXEC_B32 : SOP1_Real_gfx10_Renamed_gfx13<0x46, "s_and_not0_wrexec_b32">;
|
|
defm S_ANDN2_WREXEC_B32 : SOP1_Real_gfx10_Renamed_gfx13<0x47, "s_and_not1_wrexec_b32">;
|
|
defm S_MOVRELSD_2_B32 : SOP1_Real_gfx10_gfx13<0x049>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOP1 - GFX6, GFX7, GFX10, GFX11, GFX13
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
|
|
multiclass SOP1_Real_gfx6_gfx7<bits<8> op> {
|
|
defvar ps = !cast<SOP1_Pseudo>(NAME);
|
|
def _gfx6_gfx7 : SOP1_Real<op, ps>,
|
|
Select_gfx6_gfx7<ps.PseudoInstr>;
|
|
}
|
|
|
|
multiclass SOP1_Real_gfx6_gfx7_gfx10<bits<8> op> :
|
|
SOP1_Real_gfx6_gfx7<op>, SOP1_Real_gfx10<op>;
|
|
|
|
multiclass SOP1_Real_gfx6_gfx7_gfx10_gfx13<bits<8> op> :
|
|
SOP1_Real_gfx6_gfx7<op>, SOP1_Real_gfx10<op>, SOP1_Real_gfx13<op>;
|
|
|
|
multiclass SOP1_Real_gfx6_gfx7_gfx10_Renamed_gfx13<bits<8> op, string gfx13_name> :
|
|
SOP1_Real_gfx6_gfx7<op>, SOP1_Real_gfx10<op>, SOP1_Real_gfx13<op, gfx13_name>;
|
|
|
|
multiclass SOP1_Real_gfx6_gfx7_gfx10_gfx11_gfx12_gfx13<bits<8> op> :
|
|
SOP1_Real_gfx6_gfx7<op>, SOP1_Real_gfx10<op>, SOP1_Real_gfx11<op>,
|
|
SOP1_Real_gfx12<op>, SOP1_Real_gfx13<op>;
|
|
|
|
defm S_CBRANCH_JOIN : SOP1_Real_gfx6_gfx7<0x032>;
|
|
|
|
defm S_MOV_B32 : SOP1_Real_gfx6_gfx7_gfx10_gfx13<0x003>;
|
|
defm S_MOV_B64 : SOP1_Real_gfx6_gfx7_gfx10_gfx13<0x004>;
|
|
defm S_CMOV_B32 : SOP1_Real_gfx6_gfx7_gfx10_gfx13<0x005>;
|
|
defm S_CMOV_B64 : SOP1_Real_gfx6_gfx7_gfx10_gfx13<0x006>;
|
|
defm S_NOT_B32 : SOP1_Real_gfx6_gfx7_gfx10_gfx13<0x007>;
|
|
defm S_NOT_B64 : SOP1_Real_gfx6_gfx7_gfx10_gfx13<0x008>;
|
|
defm S_WQM_B32 : SOP1_Real_gfx6_gfx7_gfx10_gfx13<0x009>;
|
|
defm S_WQM_B64 : SOP1_Real_gfx6_gfx7_gfx10_gfx13<0x00a>;
|
|
defm S_BREV_B32 : SOP1_Real_gfx6_gfx7_gfx10_gfx13<0x00b>;
|
|
defm S_BREV_B64 : SOP1_Real_gfx6_gfx7_gfx10_gfx13<0x00c>;
|
|
defm S_BCNT0_I32_B32 : SOP1_Real_gfx6_gfx7_gfx10_gfx13<0x00d>;
|
|
defm S_BCNT0_I32_B64 : SOP1_Real_gfx6_gfx7_gfx10_gfx13<0x00e>;
|
|
defm S_BCNT1_I32_B32 : SOP1_Real_gfx6_gfx7_gfx10_gfx13<0x00f>;
|
|
defm S_BCNT1_I32_B64 : SOP1_Real_gfx6_gfx7_gfx10_gfx13<0x010>;
|
|
defm S_FF0_I32_B32 : SOP1_Real_gfx6_gfx7_gfx10<0x011>;
|
|
defm S_FF0_I32_B64 : SOP1_Real_gfx6_gfx7_gfx10<0x012>;
|
|
defm S_FF1_I32_B32 : SOP1_Real_gfx6_gfx7_gfx10_Renamed_gfx13<0x013, "s_ctz_i32_b32">;
|
|
defm S_FF1_I32_B64 : SOP1_Real_gfx6_gfx7_gfx10_Renamed_gfx13<0x014, "s_ctz_i32_b64">;
|
|
defm S_FLBIT_I32_B32 : SOP1_Real_gfx6_gfx7_gfx10_Renamed_gfx13<0x015, "s_clz_i32_u32">;
|
|
defm S_FLBIT_I32_B64 : SOP1_Real_gfx6_gfx7_gfx10_Renamed_gfx13<0x016, "s_clz_i32_u64">;
|
|
defm S_FLBIT_I32 : SOP1_Real_gfx6_gfx7_gfx10_Renamed_gfx13<0x017, "s_cls_i32">;
|
|
defm S_FLBIT_I32_I64 : SOP1_Real_gfx6_gfx7_gfx10_Renamed_gfx13<0x018, "s_cls_i32_i64">;
|
|
defm S_SEXT_I32_I8 : SOP1_Real_gfx6_gfx7_gfx10_gfx13<0x019>;
|
|
defm S_SEXT_I32_I16 : SOP1_Real_gfx6_gfx7_gfx10_gfx13<0x01a>;
|
|
defm S_BITSET0_B32 : SOP1_Real_gfx6_gfx7_gfx10_gfx13<0x01b>;
|
|
defm S_BITSET0_B64 : SOP1_Real_gfx6_gfx7_gfx10_gfx13<0x01c>;
|
|
defm S_BITSET1_B32 : SOP1_Real_gfx6_gfx7_gfx10_gfx13<0x01d>;
|
|
defm S_BITSET1_B64 : SOP1_Real_gfx6_gfx7_gfx10_gfx13<0x01e>;
|
|
defm S_GETPC_B64 : SOP1_Real_gfx6_gfx7_gfx10_Renamed_gfx13<0x01f, "s_get_pc_i64">;
|
|
defm S_SETPC_B64 : SOP1_Real_gfx6_gfx7_gfx10_Renamed_gfx13<0x020, "s_set_pc_i64">;
|
|
defm S_SWAPPC_B64 : SOP1_Real_gfx6_gfx7_gfx10_Renamed_gfx13<0x021, "s_swap_pc_i64">;
|
|
defm S_RFE_B64 : SOP1_Real_gfx6_gfx7_gfx10_Renamed_gfx13<0x022, "s_rfe_i64">;
|
|
defm S_AND_SAVEEXEC_B64 : SOP1_Real_gfx6_gfx7_gfx10_gfx13<0x024>;
|
|
defm S_OR_SAVEEXEC_B64 : SOP1_Real_gfx6_gfx7_gfx10_gfx13<0x025>;
|
|
defm S_XOR_SAVEEXEC_B64 : SOP1_Real_gfx6_gfx7_gfx10_gfx13<0x026>;
|
|
defm S_ANDN2_SAVEEXEC_B64 : SOP1_Real_gfx6_gfx7_gfx10_Renamed_gfx13<0x027, "s_and_not1_saveexec_b64">;
|
|
defm S_ORN2_SAVEEXEC_B64 : SOP1_Real_gfx6_gfx7_gfx10_Renamed_gfx13<0x028, "s_or_not1_saveexec_b64">;
|
|
defm S_NAND_SAVEEXEC_B64 : SOP1_Real_gfx6_gfx7_gfx10_gfx13<0x029>;
|
|
defm S_NOR_SAVEEXEC_B64 : SOP1_Real_gfx6_gfx7_gfx10_gfx13<0x02a>;
|
|
defm S_XNOR_SAVEEXEC_B64 : SOP1_Real_gfx6_gfx7_gfx10_gfx11_gfx12_gfx13<0x02b>;
|
|
defm S_QUADMASK_B32 : SOP1_Real_gfx6_gfx7_gfx10_gfx13<0x02c>;
|
|
defm S_QUADMASK_B64 : SOP1_Real_gfx6_gfx7_gfx10_gfx13<0x02d>;
|
|
defm S_MOVRELS_B32 : SOP1_Real_gfx6_gfx7_gfx10_gfx13<0x02e>;
|
|
defm S_MOVRELS_B64 : SOP1_Real_gfx6_gfx7_gfx10_gfx13<0x02f>;
|
|
defm S_MOVRELD_B32 : SOP1_Real_gfx6_gfx7_gfx10_gfx13<0x030>;
|
|
defm S_MOVRELD_B64 : SOP1_Real_gfx6_gfx7_gfx10_gfx13<0x031>;
|
|
defm S_ABS_I32 : SOP1_Real_gfx6_gfx7_gfx10_gfx13<0x034>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOP2 - GFX13.
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOP2_Real_gfx13<bits<7> op, string name = !tolower(NAME)> {
|
|
defvar ps = !cast<SOP2_Pseudo>(NAME);
|
|
def _gfx13 : SOP2_Real32<op, ps, name>,
|
|
Select<GFX13Gen, ps.Mnemonic>;
|
|
if !ne(ps.Mnemonic, name) then
|
|
def : MnemonicAlias<ps.Mnemonic, name>, Requires<[isGFX13Only]>;
|
|
}
|
|
|
|
defm S_PACK_HL_B32_B16 : SOP2_Real_gfx13<0x37>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOP2 - GFX12, GFX13
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOP2_Real_gfx12<bits<7> op, string name = !tolower(NAME)> {
|
|
defvar ps = !cast<SOP2_Pseudo>(NAME);
|
|
def _gfx12 : SOP2_Real32<op, ps, name>,
|
|
Select<GFX12Gen, ps.PseudoInstr>;
|
|
if !ne(ps.Mnemonic, name) then
|
|
def : AMDGPUMnemonicAlias<ps.Mnemonic, name> {
|
|
let AssemblerPredicate = isGFX12Only;
|
|
}
|
|
}
|
|
|
|
multiclass SOP2_Real_gfx12_gfx13<bits<7> op, string name = !tolower(NAME)> :
|
|
SOP2_Real_gfx12<op, name>, SOP2_Real_gfx13<op, name>;
|
|
|
|
defm S_MINIMUM_F32 : SOP2_Real_gfx12_gfx13<0x04f>;
|
|
defm S_MAXIMUM_F32 : SOP2_Real_gfx12_gfx13<0x050>;
|
|
defm S_MINIMUM_F16 : SOP2_Real_gfx12_gfx13<0x051>;
|
|
defm S_MAXIMUM_F16 : SOP2_Real_gfx12_gfx13<0x052>;
|
|
defm S_ADD_U64 : SOP2_Real_gfx12_gfx13<0x053, "s_add_nc_u64">;
|
|
defm S_SUB_U64 : SOP2_Real_gfx12_gfx13<0x054, "s_sub_nc_u64">;
|
|
defm S_MUL_U64 : SOP2_Real_gfx12_gfx13<0x055>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOP2 - GFX11, GFX12, GFX13
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOP2_Real_gfx11<bits<7> op, string name = !tolower(NAME)> {
|
|
defvar ps = !cast<SOP2_Pseudo>(NAME);
|
|
def _gfx11 : SOP2_Real32<op, ps, name>,
|
|
Select<GFX11Gen, ps.PseudoInstr>;
|
|
if !ne(ps.Mnemonic, name) then
|
|
def : AMDGPUMnemonicAlias<ps.Mnemonic, name> {
|
|
let AssemblerPredicate = isGFX11Only;
|
|
}
|
|
}
|
|
|
|
multiclass SOP2_Real_gfx11_gfx12<bits<7> op, string name = !tolower(NAME)> :
|
|
SOP2_Real_gfx11<op, name>, SOP2_Real_gfx12<op, name>;
|
|
|
|
defm S_ABSDIFF_I32 : SOP2_Real_gfx11_gfx12<0x006>;
|
|
defm S_LSHL_B32 : SOP2_Real_gfx11_gfx12<0x008>;
|
|
defm S_LSHL_B64 : SOP2_Real_gfx11_gfx12<0x009>;
|
|
defm S_LSHR_B32 : SOP2_Real_gfx11_gfx12<0x00a>;
|
|
defm S_LSHR_B64 : SOP2_Real_gfx11_gfx12<0x00b>;
|
|
defm S_ASHR_I32 : SOP2_Real_gfx11_gfx12<0x00c>;
|
|
defm S_ASHR_I64 : SOP2_Real_gfx11_gfx12<0x00d>;
|
|
defm S_LSHL1_ADD_U32 : SOP2_Real_gfx11_gfx12<0x00e>;
|
|
defm S_LSHL2_ADD_U32 : SOP2_Real_gfx11_gfx12<0x00f>;
|
|
defm S_LSHL3_ADD_U32 : SOP2_Real_gfx11_gfx12<0x010>;
|
|
defm S_LSHL4_ADD_U32 : SOP2_Real_gfx11_gfx12<0x011>;
|
|
defm S_MIN_I32 : SOP2_Real_gfx11_gfx12<0x012>;
|
|
defm S_MIN_U32 : SOP2_Real_gfx11_gfx12<0x013>;
|
|
defm S_MAX_I32 : SOP2_Real_gfx11_gfx12<0x014>;
|
|
defm S_MAX_U32 : SOP2_Real_gfx11_gfx12<0x015>;
|
|
defm S_AND_B32 : SOP2_Real_gfx11_gfx12<0x016>;
|
|
defm S_AND_B64 : SOP2_Real_gfx11_gfx12<0x017>;
|
|
defm S_OR_B32 : SOP2_Real_gfx11_gfx12<0x018>;
|
|
defm S_OR_B64 : SOP2_Real_gfx11_gfx12<0x019>;
|
|
defm S_XOR_B32 : SOP2_Real_gfx11_gfx12<0x01a>;
|
|
defm S_XOR_B64 : SOP2_Real_gfx11_gfx12<0x01b>;
|
|
defm S_NAND_B32 : SOP2_Real_gfx11_gfx12<0x01c>;
|
|
defm S_NAND_B64 : SOP2_Real_gfx11_gfx12<0x01d>;
|
|
defm S_NOR_B32 : SOP2_Real_gfx11_gfx12<0x01e>;
|
|
defm S_NOR_B64 : SOP2_Real_gfx11_gfx12<0x01f>;
|
|
defm S_XNOR_B32 : SOP2_Real_gfx11_gfx12<0x020>;
|
|
defm S_XNOR_B64 : SOP2_Real_gfx11_gfx12<0x021>;
|
|
defm S_ANDN2_B32 : SOP2_Real_gfx11_gfx12<0x022, "s_and_not1_b32">;
|
|
defm S_ANDN2_B64 : SOP2_Real_gfx11_gfx12<0x023, "s_and_not1_b64">;
|
|
defm S_ORN2_B32 : SOP2_Real_gfx11_gfx12<0x024, "s_or_not1_b32">;
|
|
defm S_ORN2_B64 : SOP2_Real_gfx11_gfx12<0x025, "s_or_not1_b64">;
|
|
defm S_BFE_U32 : SOP2_Real_gfx11_gfx12<0x026>;
|
|
defm S_BFE_I32 : SOP2_Real_gfx11_gfx12<0x027>;
|
|
defm S_BFE_U64 : SOP2_Real_gfx11_gfx12<0x028>;
|
|
defm S_BFE_I64 : SOP2_Real_gfx11_gfx12<0x029>;
|
|
defm S_BFM_B32 : SOP2_Real_gfx11_gfx12<0x02a>;
|
|
defm S_BFM_B64 : SOP2_Real_gfx11_gfx12<0x02b>;
|
|
defm S_MUL_I32 : SOP2_Real_gfx11_gfx12<0x02c>;
|
|
defm S_MUL_HI_U32 : SOP2_Real_gfx11_gfx12<0x02d>;
|
|
defm S_MUL_HI_I32 : SOP2_Real_gfx11_gfx12<0x02e>;
|
|
defm S_CSELECT_B32 : SOP2_Real_gfx11_gfx12<0x030>;
|
|
defm S_CSELECT_B64 : SOP2_Real_gfx11_gfx12<0x031>;
|
|
defm S_PACK_HL_B32_B16 : SOP2_Real_gfx11_gfx12<0x035>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOP2 - GFX1150, GFX12, GFX13
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOP2_Real_gfx11_gfx12_gfx13<bits<7> op> :
|
|
SOP2_Real_gfx11<op>, SOP2_Real_gfx12<op>, SOP2_Real_gfx13<op>;
|
|
|
|
multiclass SOP2_Real_FMAK_gfx13<bits<7> op> {
|
|
def _gfx13 : SOP2_Real64<op, !cast<SOP2_Pseudo>(NAME)>,
|
|
Select<GFX13Gen, !cast<SOP2_Pseudo>(NAME).Mnemonic>;
|
|
}
|
|
|
|
multiclass SOP2_Real_FMAK_gfx12<bits<7> op> {
|
|
def _gfx12 : SOP2_Real64<op, !cast<SOP2_Pseudo>(NAME)>,
|
|
Select<GFX12Gen, !cast<SOP2_Pseudo>(NAME).PseudoInstr>;
|
|
}
|
|
|
|
multiclass SOP2_Real_FMAK_gfx11<bits<7> op> {
|
|
def _gfx11 : SOP2_Real64<op, !cast<SOP2_Pseudo>(NAME)>,
|
|
Select<GFX11Gen, !cast<SOP2_Pseudo>(NAME).PseudoInstr>;
|
|
}
|
|
|
|
multiclass SOP2_Real_FMAK_gfx11_gfx12_gfx13<bits<7> op> :
|
|
SOP2_Real_FMAK_gfx11<op>, SOP2_Real_FMAK_gfx12<op>, SOP2_Real_FMAK_gfx13<op>;
|
|
|
|
defm S_ADD_F32 : SOP2_Real_gfx11_gfx12_gfx13<0x040>;
|
|
defm S_SUB_F32 : SOP2_Real_gfx11_gfx12_gfx13<0x041>;
|
|
defm S_MUL_F32 : SOP2_Real_gfx11_gfx12_gfx13<0x044>;
|
|
defm S_FMAAK_F32 : SOP2_Real_FMAK_gfx11_gfx12_gfx13<0x045>;
|
|
defm S_FMAMK_F32 : SOP2_Real_FMAK_gfx11_gfx12_gfx13<0x046>;
|
|
defm S_FMAC_F32 : SOP2_Real_gfx11_gfx12_gfx13<0x047>;
|
|
defm S_CVT_PK_RTZ_F16_F32 : SOP2_Real_gfx11_gfx12_gfx13<0x048>;
|
|
defm S_ADD_F16 : SOP2_Real_gfx11_gfx12_gfx13<0x049>;
|
|
defm S_SUB_F16 : SOP2_Real_gfx11_gfx12_gfx13<0x04a>;
|
|
defm S_MUL_F16 : SOP2_Real_gfx11_gfx12_gfx13<0x04d>;
|
|
defm S_FMAC_F16 : SOP2_Real_gfx11_gfx12_gfx13<0x04e>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOP2 - GFX1150, GFX12, GFX13
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOP2_Real_gfx11_Renamed_gfx12_gfx13<bits<7> op, string gfx12_gfx13_name> :
|
|
SOP2_Real_gfx11<op>, SOP2_Real_gfx12<op, gfx12_gfx13_name>,
|
|
SOP2_Real_gfx13<op, gfx12_gfx13_name>;
|
|
|
|
defm S_MIN_F32 : SOP2_Real_gfx11_Renamed_gfx12_gfx13<0x042, "s_min_num_f32">;
|
|
defm S_MAX_F32 : SOP2_Real_gfx11_Renamed_gfx12_gfx13<0x043, "s_max_num_f32">;
|
|
defm S_MIN_F16 : SOP2_Real_gfx11_Renamed_gfx12_gfx13<0x04b, "s_min_num_f16">;
|
|
defm S_MAX_F16 : SOP2_Real_gfx11_Renamed_gfx12_gfx13<0x04c, "s_max_num_f16">;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOP2 - GFX10, GFX11, GFX12, GFX13
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOP2_Real_gfx10<bits<7> op> {
|
|
defvar ps = !cast<SOP2_Pseudo>(NAME);
|
|
def _gfx10 : SOP2_Real32<op, ps>,
|
|
Select<GFX10Gen, ps.PseudoInstr>;
|
|
}
|
|
|
|
multiclass SOP2_Real_gfx10_gfx13<bits<7> op> :
|
|
SOP2_Real_gfx10<op>, SOP2_Real_gfx13<op>;
|
|
|
|
multiclass SOP2_Real_gfx10_gfx11_gfx12_gfx13<bits<7> op> :
|
|
SOP2_Real_gfx10<op>, SOP2_Real_gfx11<op>, SOP2_Real_gfx12<op>,
|
|
SOP2_Real_gfx13<op>;
|
|
|
|
defm S_LSHL1_ADD_U32 : SOP2_Real_gfx10_gfx13<0x02e>;
|
|
defm S_LSHL2_ADD_U32 : SOP2_Real_gfx10_gfx13<0x02f>;
|
|
defm S_LSHL3_ADD_U32 : SOP2_Real_gfx10_gfx13<0x030>;
|
|
defm S_LSHL4_ADD_U32 : SOP2_Real_gfx10_gfx13<0x031>;
|
|
defm S_PACK_LL_B32_B16 : SOP2_Real_gfx10_gfx11_gfx12_gfx13<0x032>;
|
|
defm S_PACK_LH_B32_B16 : SOP2_Real_gfx10_gfx11_gfx12_gfx13<0x033>;
|
|
defm S_PACK_HH_B32_B16 : SOP2_Real_gfx10_gfx11_gfx12_gfx13<0x034>;
|
|
defm S_MUL_HI_U32 : SOP2_Real_gfx10_gfx13<0x035>;
|
|
defm S_MUL_HI_I32 : SOP2_Real_gfx10_gfx13<0x036>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOP2 - GFX6, GFX7, GFX10, GFX11, GFX12, GFX13
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOP2_Real_gfx6_gfx7<bits<7> op> {
|
|
defvar ps = !cast<SOP2_Pseudo>(NAME);
|
|
def _gfx6_gfx7 : SOP2_Real32<op, ps>,
|
|
Select_gfx6_gfx7<ps.PseudoInstr>;
|
|
}
|
|
|
|
multiclass SOP2_Real_gfx6_gfx7_gfx10_gfx13<bits<7> op> :
|
|
SOP2_Real_gfx6_gfx7<op>, SOP2_Real_gfx10<op>, SOP2_Real_gfx13<op>;
|
|
|
|
multiclass SOP2_Real_gfx6_gfx7_gfx10_Renamed_gfx13<bits<7> op, string gfx13_name> :
|
|
SOP2_Real_gfx6_gfx7<op>, SOP2_Real_gfx10<op>, SOP2_Real_gfx13<op, gfx13_name>;
|
|
|
|
multiclass SOP2_Real_gfx6_gfx7_gfx10_gfx11_Renamed_gfx12_gfx13<bits<7> op, string gfx12_gfx13_name> :
|
|
SOP2_Real_gfx6_gfx7<op>, SOP2_Real_gfx10<op>, SOP2_Real_gfx11<op>,
|
|
SOP2_Real_gfx12<op, gfx12_gfx13_name>, SOP2_Real_gfx13<op, gfx12_gfx13_name>;
|
|
|
|
defm S_CBRANCH_G_FORK : SOP2_Real_gfx6_gfx7<0x02b>;
|
|
|
|
defm S_ADD_U32 : SOP2_Real_gfx6_gfx7_gfx10_gfx11_Renamed_gfx12_gfx13<0x000, "s_add_co_u32">;
|
|
defm S_SUB_U32 : SOP2_Real_gfx6_gfx7_gfx10_gfx11_Renamed_gfx12_gfx13<0x001, "s_sub_co_u32">;
|
|
defm S_ADD_I32 : SOP2_Real_gfx6_gfx7_gfx10_gfx11_Renamed_gfx12_gfx13<0x002, "s_add_co_i32">;
|
|
defm S_SUB_I32 : SOP2_Real_gfx6_gfx7_gfx10_gfx11_Renamed_gfx12_gfx13<0x003, "s_sub_co_i32">;
|
|
defm S_ADDC_U32 : SOP2_Real_gfx6_gfx7_gfx10_gfx11_Renamed_gfx12_gfx13<0x004, "s_add_co_ci_u32">;
|
|
defm S_SUBB_U32 : SOP2_Real_gfx6_gfx7_gfx10_gfx11_Renamed_gfx12_gfx13<0x005, "s_sub_co_ci_u32">;
|
|
defm S_MIN_I32 : SOP2_Real_gfx6_gfx7_gfx10_gfx13<0x006>;
|
|
defm S_MIN_U32 : SOP2_Real_gfx6_gfx7_gfx10_gfx13<0x007>;
|
|
defm S_MAX_I32 : SOP2_Real_gfx6_gfx7_gfx10_gfx13<0x008>;
|
|
defm S_MAX_U32 : SOP2_Real_gfx6_gfx7_gfx10_gfx13<0x009>;
|
|
defm S_CSELECT_B32 : SOP2_Real_gfx6_gfx7_gfx10_gfx13<0x00a>;
|
|
defm S_CSELECT_B64 : SOP2_Real_gfx6_gfx7_gfx10_gfx13<0x00b>;
|
|
defm S_AND_B32 : SOP2_Real_gfx6_gfx7_gfx10_gfx13<0x00e>;
|
|
defm S_AND_B64 : SOP2_Real_gfx6_gfx7_gfx10_gfx13<0x00f>;
|
|
defm S_OR_B32 : SOP2_Real_gfx6_gfx7_gfx10_gfx13<0x010>;
|
|
defm S_OR_B64 : SOP2_Real_gfx6_gfx7_gfx10_gfx13<0x011>;
|
|
defm S_XOR_B32 : SOP2_Real_gfx6_gfx7_gfx10_gfx13<0x012>;
|
|
defm S_XOR_B64 : SOP2_Real_gfx6_gfx7_gfx10_gfx13<0x013>;
|
|
defm S_ANDN2_B32 : SOP2_Real_gfx6_gfx7_gfx10_Renamed_gfx13<0x014, "s_and_not1_b32">;
|
|
defm S_ANDN2_B64 : SOP2_Real_gfx6_gfx7_gfx10_Renamed_gfx13<0x015, "s_and_not1_b64">;
|
|
defm S_ORN2_B32 : SOP2_Real_gfx6_gfx7_gfx10_Renamed_gfx13<0x016, "s_or_not1_b32">;
|
|
defm S_ORN2_B64 : SOP2_Real_gfx6_gfx7_gfx10_Renamed_gfx13<0x017, "s_or_not1_b64">;
|
|
defm S_NAND_B32 : SOP2_Real_gfx6_gfx7_gfx10_gfx13<0x018>;
|
|
defm S_NAND_B64 : SOP2_Real_gfx6_gfx7_gfx10_gfx13<0x019>;
|
|
defm S_NOR_B32 : SOP2_Real_gfx6_gfx7_gfx10_gfx13<0x01a>;
|
|
defm S_NOR_B64 : SOP2_Real_gfx6_gfx7_gfx10_gfx13<0x01b>;
|
|
defm S_XNOR_B32 : SOP2_Real_gfx6_gfx7_gfx10_gfx13<0x01c>;
|
|
defm S_XNOR_B64 : SOP2_Real_gfx6_gfx7_gfx10_gfx13<0x01d>;
|
|
defm S_LSHL_B32 : SOP2_Real_gfx6_gfx7_gfx10_gfx13<0x01e>;
|
|
defm S_LSHL_B64 : SOP2_Real_gfx6_gfx7_gfx10_gfx13<0x01f>;
|
|
defm S_LSHR_B32 : SOP2_Real_gfx6_gfx7_gfx10_gfx13<0x020>;
|
|
defm S_LSHR_B64 : SOP2_Real_gfx6_gfx7_gfx10_gfx13<0x021>;
|
|
defm S_ASHR_I32 : SOP2_Real_gfx6_gfx7_gfx10_gfx13<0x022>;
|
|
defm S_ASHR_I64 : SOP2_Real_gfx6_gfx7_gfx10_gfx13<0x023>;
|
|
defm S_BFM_B32 : SOP2_Real_gfx6_gfx7_gfx10_gfx13<0x024>;
|
|
defm S_BFM_B64 : SOP2_Real_gfx6_gfx7_gfx10_gfx13<0x025>;
|
|
defm S_MUL_I32 : SOP2_Real_gfx6_gfx7_gfx10_gfx13<0x026>;
|
|
defm S_BFE_U32 : SOP2_Real_gfx6_gfx7_gfx10_gfx13<0x027>;
|
|
defm S_BFE_I32 : SOP2_Real_gfx6_gfx7_gfx10_gfx13<0x028>;
|
|
defm S_BFE_U64 : SOP2_Real_gfx6_gfx7_gfx10_gfx13<0x029>;
|
|
defm S_BFE_I64 : SOP2_Real_gfx6_gfx7_gfx10_gfx13<0x02a>;
|
|
defm S_ABSDIFF_I32 : SOP2_Real_gfx6_gfx7_gfx10_gfx13<0x02c>;
|
|
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOPK - GFX10 Only
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOPK_Real32_gfx10<bits<5> op> {
|
|
defvar ps = !cast<SOPK_Pseudo>(NAME);
|
|
def _gfx10 : SOPK_Real32<op, ps>,
|
|
Select<GFX10Gen, ps.PseudoInstr>;
|
|
}
|
|
|
|
multiclass SOPK_Real64_gfx10<bits<5> op> {
|
|
defvar ps = !cast<SOPK_Pseudo>(NAME);
|
|
def _gfx10 : SOPK_Real64<op, ps>,
|
|
Select<GFX10Gen, ps.PseudoInstr>;
|
|
}
|
|
|
|
defm S_WAITCNT_VSCNT : SOPK_Real32_gfx10<0x017>;
|
|
defm S_WAITCNT_VMCNT : SOPK_Real32_gfx10<0x018>;
|
|
defm S_WAITCNT_EXPCNT : SOPK_Real32_gfx10<0x019>;
|
|
defm S_WAITCNT_LGKMCNT : SOPK_Real32_gfx10<0x01a>;
|
|
defm S_SUBVECTOR_LOOP_BEGIN : SOPK_Real32_gfx10<0x01b>;
|
|
defm S_SUBVECTOR_LOOP_END : SOPK_Real32_gfx10<0x01c>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOPK - GFX11 Only
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOPK_Real32_gfx11<bits<5> op> {
|
|
def _gfx11 : SOPK_Real32<op, !cast<SOPK_Pseudo>(NAME)>,
|
|
Select<GFX11Gen, !cast<SOPK_Pseudo>(NAME).PseudoInstr>;
|
|
}
|
|
|
|
multiclass SOPK_Real64_gfx11<bits<5> op> {
|
|
def _gfx11 : SOPK_Real64<op, !cast<SOPK_Pseudo>(NAME)>,
|
|
Select<GFX11Gen, !cast<SOPK_Pseudo>(NAME).PseudoInstr>;
|
|
}
|
|
|
|
defm S_SUBVECTOR_LOOP_BEGIN : SOPK_Real32_gfx11<0x016>;
|
|
defm S_SUBVECTOR_LOOP_END : SOPK_Real32_gfx11<0x017>;
|
|
defm S_WAITCNT_VSCNT : SOPK_Real32_gfx11<0x018>;
|
|
defm S_WAITCNT_VMCNT : SOPK_Real32_gfx11<0x019>;
|
|
defm S_WAITCNT_EXPCNT : SOPK_Real32_gfx11<0x01a>;
|
|
defm S_WAITCNT_LGKMCNT : SOPK_Real32_gfx11<0x01b>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOPK - GFX11, GFX12.
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOPK_Real32_gfx12<bits<5> op, string name = !tolower(NAME)> {
|
|
defvar ps = !cast<SOPK_Pseudo>(NAME);
|
|
def _gfx12 : SOPK_Real32<op, ps, name>,
|
|
Select<GFX12Gen, ps.PseudoInstr>;
|
|
if !ne(ps.Mnemonic, name) then
|
|
def : AMDGPUMnemonicAlias<ps.Mnemonic, name> {
|
|
let AssemblerPredicate = isGFX12Plus;
|
|
}
|
|
}
|
|
|
|
multiclass SOPK_Real64_gfx12<bits<5> op> {
|
|
def _gfx12 : SOPK_Real64<op, !cast<SOPK_Pseudo>(NAME)>,
|
|
Select<GFX12Gen, !cast<SOPK_Pseudo>(NAME).PseudoInstr>;
|
|
}
|
|
|
|
multiclass SOPK_Real32_gfx11_gfx12<bits<5> op> :
|
|
SOPK_Real32_gfx11<op>, SOPK_Real32_gfx12<op>;
|
|
|
|
multiclass SOPK_Real64_gfx11_gfx12<bits<5> op> :
|
|
SOPK_Real64_gfx11<op>, SOPK_Real64_gfx12<op>;
|
|
|
|
multiclass SOPK_Real32_gfx1250<bits<5> op, string name = !tolower(NAME)> {
|
|
defvar ps = !cast<SOPK_Pseudo>(NAME);
|
|
def _gfx1250 : SOPK_Real32<op, ps, name>,
|
|
Select<GFX1250Gen, ps.PseudoInstr>;
|
|
if !ne(ps.Mnemonic, name) then
|
|
let AssemblerPredicate = isGFX1250Plus in
|
|
def : AMDGPUMnemonicAlias<ps.Mnemonic, name>;
|
|
}
|
|
|
|
defm S_GETREG_B32 : SOPK_Real32_gfx11_gfx12<0x011>;
|
|
defm S_SETREG_B32 : SOPK_Real32_gfx11_gfx12<0x012>;
|
|
defm S_SETREG_IMM32_B32 : SOPK_Real64_gfx11_gfx12<0x013>;
|
|
let OtherPredicates = [isNotGFX1250Plus] in
|
|
defm S_CALL_B64 : SOPK_Real32_gfx11_gfx12<0x014>;
|
|
defm S_CALL_B64 : SOPK_Real32_gfx1250<0x014, "s_call_i64">;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOPK - GFX10, GFX11, GFX12, GFX13.
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOPK_Real32_gfx13<bits<5> op, string name = !tolower(NAME)> {
|
|
defvar ps = !cast<SOPK_Pseudo>(NAME);
|
|
def _gfx13 : SOPK_Real32<op, ps, name>,
|
|
Select<GFX13Gen, ps.Mnemonic>;
|
|
if !ne(ps.Mnemonic, name) then
|
|
def : MnemonicAlias<ps.Mnemonic, name>, Requires<[isGFX13Only]>;
|
|
}
|
|
|
|
multiclass SOPK_Real64_gfx13<bits<5> op> {
|
|
defvar ps = !cast<SOPK_Pseudo>(NAME);
|
|
def _gfx13 : SOPK_Real64<op, ps>,
|
|
Select<GFX13Gen, ps.Mnemonic>;
|
|
}
|
|
|
|
multiclass SOPK_Real32_gfx10_gfx11_gfx12_gfx13<bits<5> op> :
|
|
SOPK_Real32_gfx10<op>, SOPK_Real32_gfx11<op>, SOPK_Real32_gfx12<op>,
|
|
SOPK_Real32_gfx13<op>;
|
|
|
|
defm S_VERSION : SOPK_Real32_gfx10_gfx11_gfx12_gfx13<0x001>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOPK - GFX10, GFX13.
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOPK_Real32_gfx10_Renamed_gfx13<bits<5> op, string gfx13_name> :
|
|
SOPK_Real32_gfx10<op>, SOPK_Real32_gfx13<op, gfx13_name>;
|
|
|
|
defm S_CALL_B64 : SOPK_Real32_gfx10_Renamed_gfx13<0x016, "s_call_i64">;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOPK - GFX6, GFX7.
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOPK_Real32_gfx6_gfx7<bits<5> op> {
|
|
defvar ps = !cast<SOPK_Pseudo>(NAME);
|
|
def _gfx6_gfx7 : SOPK_Real32<op, ps>,
|
|
Select_gfx6_gfx7<ps.PseudoInstr>;
|
|
}
|
|
|
|
defm S_CBRANCH_I_FORK : SOPK_Real32_gfx6_gfx7<0x011>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOPK - GFX6, GFX7, GFX10, GFX11.
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOPK_Real32_gfx6_gfx7_gfx10_gfx11<bits<5> op> :
|
|
SOPK_Real32_gfx6_gfx7<op>, SOPK_Real32_gfx10<op>, SOPK_Real32_gfx11<op>;
|
|
|
|
defm S_CMPK_EQ_I32 : SOPK_Real32_gfx6_gfx7_gfx10_gfx11<0x003>;
|
|
defm S_CMPK_LG_I32 : SOPK_Real32_gfx6_gfx7_gfx10_gfx11<0x004>;
|
|
defm S_CMPK_GT_I32 : SOPK_Real32_gfx6_gfx7_gfx10_gfx11<0x005>;
|
|
defm S_CMPK_GE_I32 : SOPK_Real32_gfx6_gfx7_gfx10_gfx11<0x006>;
|
|
defm S_CMPK_LT_I32 : SOPK_Real32_gfx6_gfx7_gfx10_gfx11<0x007>;
|
|
defm S_CMPK_LE_I32 : SOPK_Real32_gfx6_gfx7_gfx10_gfx11<0x008>;
|
|
defm S_CMPK_EQ_U32 : SOPK_Real32_gfx6_gfx7_gfx10_gfx11<0x009>;
|
|
defm S_CMPK_LG_U32 : SOPK_Real32_gfx6_gfx7_gfx10_gfx11<0x00a>;
|
|
defm S_CMPK_GT_U32 : SOPK_Real32_gfx6_gfx7_gfx10_gfx11<0x00b>;
|
|
defm S_CMPK_GE_U32 : SOPK_Real32_gfx6_gfx7_gfx10_gfx11<0x00c>;
|
|
defm S_CMPK_LT_U32 : SOPK_Real32_gfx6_gfx7_gfx10_gfx11<0x00d>;
|
|
defm S_CMPK_LE_U32 : SOPK_Real32_gfx6_gfx7_gfx10_gfx11<0x00e>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOPK - GFX6, GFX7, GFX10, GFX13.
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOPK_Real64_gfx6_gfx7<bits<5> op> {
|
|
defvar ps = !cast<SOPK_Pseudo>(NAME);
|
|
def _gfx6_gfx7 : SOPK_Real64<op, ps>,
|
|
Select_gfx6_gfx7<ps.PseudoInstr>;
|
|
}
|
|
|
|
multiclass SOPK_Real32_gfx6_gfx7_gfx10_gfx13<bits<5> op> :
|
|
SOPK_Real32_gfx6_gfx7<op>, SOPK_Real32_gfx10<op>, SOPK_Real32_gfx13<op>;
|
|
|
|
multiclass SOPK_Real64_gfx6_gfx7_gfx10_gfx13<bits<5> op> :
|
|
SOPK_Real64_gfx6_gfx7<op>, SOPK_Real64_gfx10<op>, SOPK_Real64_gfx13<op>;
|
|
|
|
defm S_GETREG_B32 : SOPK_Real32_gfx6_gfx7_gfx10_gfx13<0x012>;
|
|
defm S_SETREG_B32 : SOPK_Real32_gfx6_gfx7_gfx10_gfx13<0x013>;
|
|
defm S_SETREG_IMM32_B32 : SOPK_Real64_gfx6_gfx7_gfx10_gfx13<0x015>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOPK - GFX6, GFX7, GFX10, GFX11, GFX12, GFX13.
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOPK_Real32_gfx6_gfx7_gfx10_gfx11_gfx12_gfx13<bits<5> op> :
|
|
SOPK_Real32_gfx6_gfx7<op>, SOPK_Real32_gfx10<op>, SOPK_Real32_gfx11<op>,
|
|
SOPK_Real32_gfx12<op>, SOPK_Real32_gfx13<op>;
|
|
|
|
multiclass SOPK_Real32_gfx6_gfx7_gfx10_gfx11_Renamed_gfx12_gfx13<bits<5> op, string gfx12_gfx13_name> :
|
|
SOPK_Real32_gfx6_gfx7<op>, SOPK_Real32_gfx10<op>, SOPK_Real32_gfx11<op>,
|
|
SOPK_Real32_gfx12<op, gfx12_gfx13_name>, SOPK_Real32_gfx13<op, gfx12_gfx13_name>;
|
|
|
|
defm S_MOVK_I32 : SOPK_Real32_gfx6_gfx7_gfx10_gfx11_gfx12_gfx13<0x000>;
|
|
defm S_CMOVK_I32 : SOPK_Real32_gfx6_gfx7_gfx10_gfx11_gfx12_gfx13<0x002>;
|
|
defm S_ADDK_I32 : SOPK_Real32_gfx6_gfx7_gfx10_gfx11_Renamed_gfx12_gfx13<0x00f, "s_addk_co_i32">;
|
|
defm S_MULK_I32 : SOPK_Real32_gfx6_gfx7_gfx10_gfx11_gfx12_gfx13<0x010>;
|
|
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOPP - GFX13 only
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOPP_Real_32_gfx13<bits<7> op, string name = !cast<SOPP_Pseudo>(NAME).Mnemonic, bit compat_alias = 1> {
|
|
defvar ps = !cast<SOPP_Pseudo>(NAME);
|
|
def _gfx13 : SOPP_Real_32<op, ps, name>,
|
|
Select<GFX13Gen, ps.Mnemonic>,
|
|
SOPPRelaxTable<0, ps.KeyName, "_gfx13">;
|
|
if !and(compat_alias, !ne(ps.Mnemonic, name)) then
|
|
def : MnemonicAlias<ps.Mnemonic, name>, Requires<[isGFX13Only]>;
|
|
}
|
|
|
|
multiclass SOPP_Real_64_gfx13<bits<7> op> {
|
|
defvar ps = !cast<SOPP_Pseudo>(NAME);
|
|
def _gfx13 : SOPP_Real_64<op, ps, ps.Mnemonic>,
|
|
Select<GFX13Gen, ps.Mnemonic>,
|
|
SOPPRelaxTable<1, ps.KeyName, "_gfx13">;
|
|
}
|
|
|
|
defm S_WAKEUP_imm : SOPP_Real_32_gfx13<0x003>;
|
|
defm S_BARRIER_WAIT : SOPP_Real_32_gfx13<0x2b>;
|
|
defm S_MONITOR_SLEEP : SOPP_Real_32_gfx13<0x2c>;
|
|
defm S_DELAY_ALU : SOPP_Real_32_gfx13<0x2e>;
|
|
defm S_WAIT_EVENT : SOPP_Real_32_gfx13<0x2f>;
|
|
defm S_BARRIER_LEAVE_IMM : SOPP_Real_32_gfx13<0x31>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOPP - GFX12 only.
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOPP_Real_32_gfx12<bits<7> op, string name = !tolower(NAME)> {
|
|
defvar ps = !cast<SOPP_Pseudo>(NAME);
|
|
def _gfx12 : SOPP_Real_32<op, ps, name>,
|
|
Select<GFX12Gen, ps.PseudoInstr>;
|
|
if !ne(ps.Mnemonic, name) then
|
|
def : AMDGPUMnemonicAlias<ps.Mnemonic, name> {
|
|
let AssemblerPredicate = isGFX12Only;
|
|
}
|
|
}
|
|
|
|
multiclass SOPP_Real_64_gfx12<bits<7> op> {
|
|
def _gfx12 : SOPP_Real_64<op, !cast<SOPP_Pseudo>(NAME), !cast<SOPP_Pseudo>(NAME).Mnemonic>,
|
|
Select<GFX12Gen, !cast<SOPP_Pseudo>(NAME).PseudoInstr>,
|
|
SOPPRelaxTable<1, !cast<SOPP_Pseudo>(NAME).KeyName, "_gfx12">;
|
|
}
|
|
|
|
defm S_BARRIER_WAIT : SOPP_Real_32_gfx12<0x014>;
|
|
defm S_BARRIER_LEAVE : SOPP_Real_32_gfx12<0x015>;
|
|
defm S_WAIT_BVHCNT : SOPP_Real_32_gfx12<0x043>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOPP - GFX11 only.
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOPP_Real_32_gfx11<bits<7> op, string name = !tolower(NAME)> {
|
|
defvar ps = !cast<SOPP_Pseudo>(NAME);
|
|
def _gfx11 : SOPP_Real_32<op, ps, name>,
|
|
Select<GFX11Gen, ps.PseudoInstr>,
|
|
SOPPRelaxTable<0, ps.KeyName, "_gfx11">;
|
|
if !ne(ps.Mnemonic, name) then
|
|
def : AMDGPUMnemonicAlias<ps.Mnemonic, name> {
|
|
let AssemblerPredicate = isGFX11Only;
|
|
}
|
|
}
|
|
|
|
multiclass SOPP_Real_64_gfx11<bits<7> op> {
|
|
def _gfx11 : SOPP_Real_64<op, !cast<SOPP_Pseudo>(NAME), !cast<SOPP_Pseudo>(NAME).Mnemonic>,
|
|
Select<GFX11Gen, !cast<SOPP_Pseudo>(NAME).PseudoInstr>,
|
|
SOPPRelaxTable<1, !cast<SOPP_Pseudo>(NAME).KeyName, "_gfx11">;
|
|
}
|
|
|
|
multiclass SOPP_Real_With_Relaxation_gfx11<bits<7> op> {
|
|
defm "" : SOPP_Real_32_gfx11<op>;
|
|
let isCodeGenOnly = 1 in
|
|
defm _pad_s_nop : SOPP_Real_64_gfx11<op>;
|
|
}
|
|
|
|
defm S_INST_PREFETCH : SOPP_Real_32_gfx11<0x004, "s_set_inst_prefetch_distance">;
|
|
defm S_CBRANCH_CDBGSYS : SOPP_Real_With_Relaxation_gfx11<0x027>;
|
|
defm S_CBRANCH_CDBGUSER : SOPP_Real_With_Relaxation_gfx11<0x028>;
|
|
defm S_CBRANCH_CDBGSYS_OR_USER : SOPP_Real_With_Relaxation_gfx11<0x029>;
|
|
defm S_CBRANCH_CDBGSYS_AND_USER : SOPP_Real_With_Relaxation_gfx11<0x02a>;
|
|
defm S_ENDPGM_ORDERED_PS_DONE : SOPP_Real_32_gfx11<0x032>;
|
|
defm S_BARRIER : SOPP_Real_32_gfx11<0x03d>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOPP - GFX10 only.
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOPP_Real_32_gfx10<bits<7> op> {
|
|
defvar ps = !cast<SOPP_Pseudo>(NAME);
|
|
def _gfx10 : SOPP_Real_32<op, ps>,
|
|
Select<GFX10Gen, ps.PseudoInstr>,
|
|
SOPPRelaxTable<0, ps.KeyName, "_gfx10">;
|
|
}
|
|
|
|
defm S_INST_PREFETCH : SOPP_Real_32_gfx10<0x020>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOPP - GFX12, GFX13.
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOPP_Real_32_gfx12_gfx13<bits<7> op> :
|
|
SOPP_Real_32_gfx12<op>, SOPP_Real_32_gfx13<op>;
|
|
|
|
defm S_WAIT_LOADCNT : SOPP_Real_32_gfx12_gfx13<0x040>;
|
|
defm S_WAIT_STORECNT : SOPP_Real_32_gfx12_gfx13<0x041>;
|
|
defm S_WAIT_SAMPLECNT : SOPP_Real_32_gfx12_gfx13<0x042>;
|
|
defm S_WAIT_EXPCNT : SOPP_Real_32_gfx12_gfx13<0x044>;
|
|
defm S_WAIT_DSCNT : SOPP_Real_32_gfx12_gfx13<0x046>;
|
|
defm S_WAIT_KMCNT : SOPP_Real_32_gfx12_gfx13<0x047>;
|
|
defm S_WAIT_LOADCNT_DSCNT : SOPP_Real_32_gfx12_gfx13<0x048>;
|
|
defm S_WAIT_STORECNT_DSCNT : SOPP_Real_32_gfx12_gfx13<0x049>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOPP - GFX1250 only.
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
defm S_MONITOR_SLEEP : SOPP_Real_32_gfx12<0x004>;
|
|
defm S_SET_VGPR_MSB : SOPP_Real_32_gfx12<0x006>;
|
|
defm S_WAIT_XCNT : SOPP_Real_32_gfx12<0x045>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOPP - GFX1250, GFX13
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
defm S_SETPRIO_INC_WG : SOPP_Real_32_gfx12_gfx13<0x03e>;
|
|
defm S_WAIT_ASYNCCNT : SOPP_Real_32_gfx12_gfx13<0x04a>;
|
|
defm S_WAIT_TENSORCNT : SOPP_Real_32_gfx12_gfx13<0x04b>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOPP - GFX10, GFX13
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOPP_Real_32_gfx10_gfx13<bits<7> op> :
|
|
SOPP_Real_32_gfx10<op>, SOPP_Real_32_gfx13<op>;
|
|
|
|
multiclass SOPP_Real_32_gfx10_Renamed_gfx13<bits<7> op, string gfx13_name> :
|
|
SOPP_Real_32_gfx10<op>, SOPP_Real_32_gfx13<op, gfx13_name>;
|
|
|
|
defm S_CLAUSE : SOPP_Real_32_gfx10_gfx13<0x021>;
|
|
defm S_WAIT_IDLE : SOPP_Real_32_gfx10_gfx13<0x022>;
|
|
defm S_ROUND_MODE : SOPP_Real_32_gfx10_gfx13<0x024>;
|
|
defm S_DENORM_MODE : SOPP_Real_32_gfx10_gfx13<0x025>;
|
|
defm S_TTRACEDATA_IMM : SOPP_Real_32_gfx10_gfx13<0x028>;
|
|
defm S_WAITCNT_DEPCTR : SOPP_Real_32_gfx10_Renamed_gfx13<0x023, "s_wait_alu">;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOPP - GFX8, GFX9.
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOPP_Real_32_gfx8_gfx9<bits<7> op> {
|
|
defvar ps = !cast<SOPP_Pseudo>(NAME);
|
|
def _vi : SOPP_Real_32<op, ps>,
|
|
Select_vi<ps.PseudoInstr>,
|
|
SOPPRelaxTable<0, ps.KeyName, "_vi">;
|
|
}
|
|
|
|
defm S_SET_GPR_IDX_OFF : SOPP_Real_32_gfx8_gfx9<0x01c>;
|
|
defm S_SET_GPR_IDX_MODE : SOPP_Real_32_gfx8_gfx9<0x01d>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOPP - GFX6, GFX7, GFX8, GFX9, GFX10.
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOPP_Real_32_gfx6_gfx7<bits<7> op> {
|
|
defvar ps = !cast<SOPP_Pseudo>(NAME);
|
|
def _gfx6_gfx7 : SOPP_Real_32<op, ps, !cast<SOPP_Pseudo>(NAME).Mnemonic>,
|
|
Select_gfx6_gfx7<ps.PseudoInstr>,
|
|
SOPPRelaxTable<0, ps.KeyName, "_gfx6_gfx7">;
|
|
}
|
|
|
|
multiclass SOPP_Real_32_gfx6_gfx7_gfx8_gfx9<bits<7> op> :
|
|
SOPP_Real_32_gfx6_gfx7<op>, SOPP_Real_32_gfx8_gfx9<op>;
|
|
|
|
multiclass SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10<bits<7> op> :
|
|
SOPP_Real_32_gfx6_gfx7_gfx8_gfx9<op>, SOPP_Real_32_gfx10<op>;
|
|
|
|
defm S_BARRIER : SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10<0x00a>;
|
|
defm S_WAITCNT : SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10<0x00c>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOPP - GFX6, GFX7, GFX8, GFX9, GFX10, GFX11, GFX12, GFX13.
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10_gfx11_gfx12_gfx13<bits<7> op> :
|
|
SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10<op>, SOPP_Real_32_gfx11<op>,
|
|
SOPP_Real_32_gfx12<op>, SOPP_Real_32_gfx13<op>;
|
|
|
|
defm S_NOP : SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10_gfx11_gfx12_gfx13<0x000>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOPP - GFX6, GFX7, GFX8, GFX9, GFX10, GFX13.
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10_gfx13<bits<7> op> :
|
|
SOPP_Real_32_gfx6_gfx7_gfx8_gfx9<op>, SOPP_Real_32_gfx10<op>,
|
|
SOPP_Real_32_gfx13<op>;
|
|
|
|
//64 bit encodings, for Relaxation
|
|
multiclass SOPP_Real_64_gfx6_gfx7<bits<7> op> {
|
|
defvar ps = !cast<SOPP_Pseudo>(NAME);
|
|
def _gfx6_gfx7 : SOPP_Real_64<op, ps>,
|
|
Select_gfx6_gfx7<ps.PseudoInstr>,
|
|
SOPPRelaxTable<1, ps.KeyName, "_gfx6_gfx7">;
|
|
}
|
|
|
|
multiclass SOPP_Real_64_gfx8_gfx9<bits<7> op> {
|
|
defvar ps = !cast<SOPP_Pseudo>(NAME);
|
|
def _vi : SOPP_Real_64<op, ps>,
|
|
Select_vi<ps.PseudoInstr>,
|
|
SOPPRelaxTable<1, ps.KeyName, "_vi">;
|
|
}
|
|
|
|
multiclass SOPP_Real_64_gfx10<bits<7> op> {
|
|
defvar ps = !cast<SOPP_Pseudo>(NAME);
|
|
def _gfx10 : SOPP_Real_64<op, ps>,
|
|
Select<GFX10Gen, ps.PseudoInstr>,
|
|
SOPPRelaxTable<1, ps.KeyName, "_gfx10">;
|
|
}
|
|
|
|
multiclass SOPP_Real_64_gfx6_gfx7_gfx8_gfx9<bits<7> op> :
|
|
SOPP_Real_64_gfx6_gfx7<op>, SOPP_Real_64_gfx8_gfx9<op>;
|
|
|
|
multiclass SOPP_Real_64_gfx6_gfx7_gfx8_gfx9_gfx10_gfx13<bits<7> op> :
|
|
SOPP_Real_64_gfx6_gfx7_gfx8_gfx9<op>, SOPP_Real_64_gfx10<op>,
|
|
SOPP_Real_64_gfx13<op>;
|
|
|
|
multiclass SOPP_Real_With_Relaxation_gfx6_gfx7_gfx8_gfx9_gfx10_gfx13<bits<7> op> {
|
|
defm "" : SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10_gfx13<op>;
|
|
let isCodeGenOnly = 1 in
|
|
defm _pad_s_nop : SOPP_Real_64_gfx6_gfx7_gfx8_gfx9_gfx10_gfx13<op>;
|
|
}
|
|
|
|
defm S_ENDPGM : SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10_gfx13<0x001>;
|
|
defm S_SETHALT : SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10_gfx13<0x00d>;
|
|
defm S_SETKILL : SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10_gfx13<0x00b>;
|
|
defm S_SLEEP : SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10_gfx13<0x00e>;
|
|
defm S_SETPRIO : SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10_gfx13<0x00f>;
|
|
defm S_SENDMSG : SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10_gfx13<0x010>;
|
|
defm S_SENDMSGHALT : SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10_gfx13<0x011>;
|
|
defm S_TRAP : SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10_gfx13<0x012>;
|
|
defm S_ICACHE_INV : SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10_gfx13<0x013>;
|
|
defm S_INCPERFLEVEL : SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10_gfx13<0x014>;
|
|
defm S_DECPERFLEVEL : SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10_gfx13<0x015>;
|
|
defm S_TTRACEDATA : SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10_gfx13<0x016>;
|
|
defm S_ENDPGM_SAVED : SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10_gfx13<0x01B>;
|
|
|
|
let isBranch = 1 in {
|
|
defm S_BRANCH : SOPP_Real_With_Relaxation_gfx6_gfx7_gfx8_gfx9_gfx10_gfx13<0x002>;
|
|
defm S_CBRANCH_SCC0 : SOPP_Real_With_Relaxation_gfx6_gfx7_gfx8_gfx9_gfx10_gfx13<0x004>;
|
|
defm S_CBRANCH_SCC1 : SOPP_Real_With_Relaxation_gfx6_gfx7_gfx8_gfx9_gfx10_gfx13<0x005>;
|
|
defm S_CBRANCH_VCCZ : SOPP_Real_With_Relaxation_gfx6_gfx7_gfx8_gfx9_gfx10_gfx13<0x006>;
|
|
defm S_CBRANCH_VCCNZ : SOPP_Real_With_Relaxation_gfx6_gfx7_gfx8_gfx9_gfx10_gfx13<0x007>;
|
|
defm S_CBRANCH_EXECZ : SOPP_Real_With_Relaxation_gfx6_gfx7_gfx8_gfx9_gfx10_gfx13<0x008>;
|
|
defm S_CBRANCH_EXECNZ : SOPP_Real_With_Relaxation_gfx6_gfx7_gfx8_gfx9_gfx10_gfx13<0x009>;
|
|
}
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOPP - GFX6, GFX7, GFX8, GFX9, GFX10.
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOPP_Real_64_gfx6_gfx7_gfx8_gfx9_gfx10<bits<7> op> :
|
|
SOPP_Real_64_gfx6_gfx7_gfx8_gfx9<op>, SOPP_Real_64_gfx10<op>;
|
|
|
|
//relaxation for insts with no operands not implemented
|
|
multiclass SOPP_Real_With_Relaxation_gfx6_gfx7_gfx8_gfx9_gfx10<bits<7> op> {
|
|
defm "" : SOPP_Real_32_gfx6_gfx7_gfx8_gfx9_gfx10<op>;
|
|
let isCodeGenOnly = 1 in
|
|
defm _pad_s_nop : SOPP_Real_64_gfx6_gfx7_gfx8_gfx9_gfx10<op>;
|
|
}
|
|
|
|
let isBranch = 1 in {
|
|
defm S_CBRANCH_CDBGSYS : SOPP_Real_With_Relaxation_gfx6_gfx7_gfx8_gfx9_gfx10<0x017>;
|
|
defm S_CBRANCH_CDBGUSER : SOPP_Real_With_Relaxation_gfx6_gfx7_gfx8_gfx9_gfx10<0x018>;
|
|
defm S_CBRANCH_CDBGSYS_OR_USER : SOPP_Real_With_Relaxation_gfx6_gfx7_gfx8_gfx9_gfx10<0x019>;
|
|
defm S_CBRANCH_CDBGSYS_AND_USER : SOPP_Real_With_Relaxation_gfx6_gfx7_gfx8_gfx9_gfx10<0x01A>;
|
|
}
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOPP - GFX10, GFX11, GFX12, GFX13.
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOPP_Real_32_gfx10_gfx11_gfx12_gfx13<bits<7> op> :
|
|
SOPP_Real_32_gfx10<op>, SOPP_Real_32_gfx11<op>, SOPP_Real_32_gfx12<op>,
|
|
SOPP_Real_32_gfx13<op>;
|
|
|
|
defm S_CODE_END : SOPP_Real_32_gfx10_gfx11_gfx12_gfx13<0x01f>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOPP - GFX11, GFX12.
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOPP_Real_32_gfx11_gfx12<bits<7> op> :
|
|
SOPP_Real_32_gfx11<op>, SOPP_Real_32_gfx12<op>;
|
|
|
|
multiclass SOPP_Real_32_gfx11_Renamed_gfx12<bits<7> op, string gfx12_name> :
|
|
SOPP_Real_32_gfx11<op>, SOPP_Real_32_gfx12<op, gfx12_name>;
|
|
|
|
multiclass SOPP_Real_With_Relaxation_gfx12<bits<7> op> {
|
|
defm "" : SOPP_Real_32_gfx12<op>;
|
|
let isCodeGenOnly = 1 in
|
|
defm _pad_s_nop : SOPP_Real_64_gfx12<op>;
|
|
}
|
|
|
|
multiclass SOPP_Real_With_Relaxation_gfx11_gfx12<bits<7>op> :
|
|
SOPP_Real_With_Relaxation_gfx11<op>, SOPP_Real_With_Relaxation_gfx12<op>;
|
|
|
|
defm S_SETKILL : SOPP_Real_32_gfx11_gfx12<0x001>;
|
|
defm S_SETHALT : SOPP_Real_32_gfx11_gfx12<0x002>;
|
|
defm S_SLEEP : SOPP_Real_32_gfx11_gfx12<0x003>;
|
|
defm S_CLAUSE : SOPP_Real_32_gfx11_gfx12<0x005>;
|
|
defm S_DELAY_ALU : SOPP_Real_32_gfx11_gfx12<0x007>;
|
|
defm S_WAITCNT_DEPCTR : SOPP_Real_32_gfx11_Renamed_gfx12<0x008, "s_wait_alu">;
|
|
defm S_WAITCNT : SOPP_Real_32_gfx11_gfx12<0x009>;
|
|
defm S_WAIT_IDLE : SOPP_Real_32_gfx11_gfx12<0x00a>;
|
|
defm S_WAIT_EVENT : SOPP_Real_32_gfx11_gfx12<0x00b>;
|
|
defm S_TRAP : SOPP_Real_32_gfx11_gfx12<0x010>;
|
|
defm S_ROUND_MODE : SOPP_Real_32_gfx11_gfx12<0x011>;
|
|
defm S_DENORM_MODE : SOPP_Real_32_gfx11_gfx12<0x012>;
|
|
defm S_BRANCH : SOPP_Real_With_Relaxation_gfx11_gfx12<0x020>;
|
|
defm S_CBRANCH_SCC0 : SOPP_Real_With_Relaxation_gfx11_gfx12<0x021>;
|
|
defm S_CBRANCH_SCC1 : SOPP_Real_With_Relaxation_gfx11_gfx12<0x022>;
|
|
defm S_CBRANCH_VCCZ : SOPP_Real_With_Relaxation_gfx11_gfx12<0x023>;
|
|
defm S_CBRANCH_VCCNZ : SOPP_Real_With_Relaxation_gfx11_gfx12<0x024>;
|
|
defm S_CBRANCH_EXECZ : SOPP_Real_With_Relaxation_gfx11_gfx12<0x025>;
|
|
defm S_CBRANCH_EXECNZ : SOPP_Real_With_Relaxation_gfx11_gfx12<0x026>;
|
|
defm S_ENDPGM : SOPP_Real_32_gfx11_gfx12<0x030>;
|
|
defm S_ENDPGM_SAVED : SOPP_Real_32_gfx11_gfx12<0x031>;
|
|
defm S_WAKEUP : SOPP_Real_32_gfx11_gfx12<0x034>;
|
|
defm S_SETPRIO : SOPP_Real_32_gfx11_gfx12<0x035>;
|
|
defm S_SENDMSG : SOPP_Real_32_gfx11_gfx12<0x036>;
|
|
defm S_SENDMSGHALT : SOPP_Real_32_gfx11_gfx12<0x037>;
|
|
defm S_INCPERFLEVEL : SOPP_Real_32_gfx11_gfx12<0x038>;
|
|
defm S_DECPERFLEVEL : SOPP_Real_32_gfx11_gfx12<0x039>;
|
|
defm S_TTRACEDATA : SOPP_Real_32_gfx11_gfx12<0x03a>;
|
|
defm S_TTRACEDATA_IMM : SOPP_Real_32_gfx11_gfx12<0x03b>;
|
|
defm S_ICACHE_INV : SOPP_Real_32_gfx11_gfx12<0x03c>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOPP - GFX8, GFX9, GFX10.
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOPP_Real_32_gfx8_gfx9_gfx10<bits<7> op> :
|
|
SOPP_Real_32_gfx8_gfx9<op>, SOPP_Real_32_gfx10<op>;
|
|
|
|
defm S_WAKEUP : SOPP_Real_32_gfx8_gfx9_gfx10<0x003>;
|
|
defm S_ENDPGM_ORDERED_PS_DONE : SOPP_Real_32_gfx8_gfx9_gfx10<0x01e>;
|
|
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOPC - GFX11, GFX12.
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOPC_Real_gfx12<bits<7> op> {
|
|
def _gfx12 : SOPC_Real<op, !cast<SOPC_Pseudo>(NAME)>,
|
|
Select<GFX12Gen, !cast<SOPC_Pseudo>(NAME).PseudoInstr>;
|
|
}
|
|
|
|
multiclass SOPC_Real_gfx11<bits<7> op> {
|
|
def _gfx11 : SOPC_Real<op, !cast<SOPC_Pseudo>(NAME)>,
|
|
Select<GFX11Gen, !cast<SOPC_Pseudo>(NAME).PseudoInstr>;
|
|
}
|
|
|
|
multiclass SOPC_Real_gfx11_gfx12<bits<7> op> :
|
|
SOPC_Real_gfx11<op>, SOPC_Real_gfx12<op>;
|
|
|
|
defm S_CMP_EQ_U64 : SOPC_Real_gfx11_gfx12<0x10>;
|
|
defm S_CMP_LG_U64 : SOPC_Real_gfx11_gfx12<0x11>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOPC - GFX1150, GFX12, GFX13
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOPC_Real_gfx13<bits<7> op> {
|
|
def _gfx13 : SOPC_Real<op, !cast<SOPC_Pseudo>(NAME)>,
|
|
Select<GFX13Gen, !cast<SOPC_Pseudo>(NAME).Mnemonic>;
|
|
}
|
|
|
|
multiclass SOPC_Real_gfx11_gfx12_gfx13<bits<7> op> :
|
|
SOPC_Real_gfx11<op>, SOPC_Real_gfx12<op>, SOPC_Real_gfx13<op>;
|
|
|
|
defm S_CMP_LT_F32 : SOPC_Real_gfx11_gfx12_gfx13<0x41>;
|
|
defm S_CMP_EQ_F32 : SOPC_Real_gfx11_gfx12_gfx13<0x42>;
|
|
defm S_CMP_LE_F32 : SOPC_Real_gfx11_gfx12_gfx13<0x43>;
|
|
defm S_CMP_GT_F32 : SOPC_Real_gfx11_gfx12_gfx13<0x44>;
|
|
defm S_CMP_LG_F32 : SOPC_Real_gfx11_gfx12_gfx13<0x45>;
|
|
defm S_CMP_GE_F32 : SOPC_Real_gfx11_gfx12_gfx13<0x46>;
|
|
defm S_CMP_O_F32 : SOPC_Real_gfx11_gfx12_gfx13<0x47>;
|
|
defm S_CMP_U_F32 : SOPC_Real_gfx11_gfx12_gfx13<0x48>;
|
|
defm S_CMP_NGE_F32 : SOPC_Real_gfx11_gfx12_gfx13<0x49>;
|
|
defm S_CMP_NLG_F32 : SOPC_Real_gfx11_gfx12_gfx13<0x4a>;
|
|
defm S_CMP_NGT_F32 : SOPC_Real_gfx11_gfx12_gfx13<0x4b>;
|
|
defm S_CMP_NLE_F32 : SOPC_Real_gfx11_gfx12_gfx13<0x4c>;
|
|
defm S_CMP_NEQ_F32 : SOPC_Real_gfx11_gfx12_gfx13<0x4d>;
|
|
defm S_CMP_NLT_F32 : SOPC_Real_gfx11_gfx12_gfx13<0x4e>;
|
|
|
|
defm S_CMP_LT_F16 : SOPC_Real_gfx11_gfx12_gfx13<0x51>;
|
|
defm S_CMP_EQ_F16 : SOPC_Real_gfx11_gfx12_gfx13<0x52>;
|
|
defm S_CMP_LE_F16 : SOPC_Real_gfx11_gfx12_gfx13<0x53>;
|
|
defm S_CMP_GT_F16 : SOPC_Real_gfx11_gfx12_gfx13<0x54>;
|
|
defm S_CMP_LG_F16 : SOPC_Real_gfx11_gfx12_gfx13<0x55>;
|
|
defm S_CMP_GE_F16 : SOPC_Real_gfx11_gfx12_gfx13<0x56>;
|
|
defm S_CMP_O_F16 : SOPC_Real_gfx11_gfx12_gfx13<0x57>;
|
|
defm S_CMP_U_F16 : SOPC_Real_gfx11_gfx12_gfx13<0x58>;
|
|
defm S_CMP_NGE_F16 : SOPC_Real_gfx11_gfx12_gfx13<0x59>;
|
|
defm S_CMP_NLG_F16 : SOPC_Real_gfx11_gfx12_gfx13<0x5a>;
|
|
defm S_CMP_NGT_F16 : SOPC_Real_gfx11_gfx12_gfx13<0x5b>;
|
|
defm S_CMP_NLE_F16 : SOPC_Real_gfx11_gfx12_gfx13<0x5c>;
|
|
defm S_CMP_NEQ_F16 : SOPC_Real_gfx11_gfx12_gfx13<0x5d>;
|
|
defm S_CMP_NLT_F16 : SOPC_Real_gfx11_gfx12_gfx13<0x5e>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOPC - GFX8, GFX9.
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOPC_Real_gfx8_gfx9<bits<7> op> {
|
|
defvar ps = !cast<SOPC_Pseudo>(NAME);
|
|
def _vi : SOPC_Real<op, ps>,
|
|
Select_vi<ps.PseudoInstr>;
|
|
}
|
|
|
|
defm S_SET_GPR_IDX_ON : SOPC_Real_gfx8_gfx9<0x11>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOPC - GFX6, GFX7, GFX8, GFX9.
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOPC_Real_gfx6_gfx7<bits<7> op> {
|
|
defvar ps = !cast<SOPC_Pseudo>(NAME);
|
|
def _gfx6_gfx7 : SOPC_Real<op, ps>,
|
|
Select_gfx6_gfx7<ps.PseudoInstr>;
|
|
}
|
|
|
|
multiclass SOPC_Real_gfx6_gfx7_gfx8_gfx9<bits<7> op> :
|
|
SOPC_Real_gfx6_gfx7<op>, SOPC_Real_gfx8_gfx9<op>;
|
|
|
|
defm S_SETVSKIP : SOPC_Real_gfx6_gfx7_gfx8_gfx9<0x10>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOPC - GFX6, GFX7, GFX8, GFX9, GFX10, GFX13
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOPC_Real_gfx10<bits<7> op> {
|
|
defvar ps = !cast<SOPC_Pseudo>(NAME);
|
|
def _gfx10 : SOPC_Real<op, ps>,
|
|
Select<GFX10Gen, ps.PseudoInstr>;
|
|
}
|
|
|
|
multiclass SOPC_Real_gfx8_gfx9_gfx10_gfx13<bits<7> op> :
|
|
SOPC_Real_gfx8_gfx9<op>, SOPC_Real_gfx10<op>, SOPC_Real_gfx13<op>;
|
|
|
|
defm S_CMP_EQ_U64 : SOPC_Real_gfx8_gfx9_gfx10_gfx13<0x12>;
|
|
defm S_CMP_LG_U64 : SOPC_Real_gfx8_gfx9_gfx10_gfx13<0x13>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOPC - GFX6, GFX7, GFX8, GFX9, GFX10, GFX11, GFX12, GFX13
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass SOPC_Real_gfx6_gfx7_gfx8_gfx9_gfx10_gfx11_gfx12_gfx13<bits<7> op> :
|
|
SOPC_Real_gfx6_gfx7_gfx8_gfx9<op>, SOPC_Real_gfx10<op>, SOPC_Real_gfx11<op>,
|
|
SOPC_Real_gfx12<op>, SOPC_Real_gfx13<op>;
|
|
|
|
defm S_CMP_EQ_I32 : SOPC_Real_gfx6_gfx7_gfx8_gfx9_gfx10_gfx11_gfx12_gfx13<0x00>;
|
|
defm S_CMP_LG_I32 : SOPC_Real_gfx6_gfx7_gfx8_gfx9_gfx10_gfx11_gfx12_gfx13<0x01>;
|
|
defm S_CMP_GT_I32 : SOPC_Real_gfx6_gfx7_gfx8_gfx9_gfx10_gfx11_gfx12_gfx13<0x02>;
|
|
defm S_CMP_GE_I32 : SOPC_Real_gfx6_gfx7_gfx8_gfx9_gfx10_gfx11_gfx12_gfx13<0x03>;
|
|
defm S_CMP_LT_I32 : SOPC_Real_gfx6_gfx7_gfx8_gfx9_gfx10_gfx11_gfx12_gfx13<0x04>;
|
|
defm S_CMP_LE_I32 : SOPC_Real_gfx6_gfx7_gfx8_gfx9_gfx10_gfx11_gfx12_gfx13<0x05>;
|
|
defm S_CMP_EQ_U32 : SOPC_Real_gfx6_gfx7_gfx8_gfx9_gfx10_gfx11_gfx12_gfx13<0x06>;
|
|
defm S_CMP_LG_U32 : SOPC_Real_gfx6_gfx7_gfx8_gfx9_gfx10_gfx11_gfx12_gfx13<0x07>;
|
|
defm S_CMP_GT_U32 : SOPC_Real_gfx6_gfx7_gfx8_gfx9_gfx10_gfx11_gfx12_gfx13<0x08>;
|
|
defm S_CMP_GE_U32 : SOPC_Real_gfx6_gfx7_gfx8_gfx9_gfx10_gfx11_gfx12_gfx13<0x09>;
|
|
defm S_CMP_LT_U32 : SOPC_Real_gfx6_gfx7_gfx8_gfx9_gfx10_gfx11_gfx12_gfx13<0x0a>;
|
|
defm S_CMP_LE_U32 : SOPC_Real_gfx6_gfx7_gfx8_gfx9_gfx10_gfx11_gfx12_gfx13<0x0b>;
|
|
defm S_BITCMP0_B32 : SOPC_Real_gfx6_gfx7_gfx8_gfx9_gfx10_gfx11_gfx12_gfx13<0x0c>;
|
|
defm S_BITCMP1_B32 : SOPC_Real_gfx6_gfx7_gfx8_gfx9_gfx10_gfx11_gfx12_gfx13<0x0d>;
|
|
defm S_BITCMP0_B64 : SOPC_Real_gfx6_gfx7_gfx8_gfx9_gfx10_gfx11_gfx12_gfx13<0x0e>;
|
|
defm S_BITCMP1_B64 : SOPC_Real_gfx6_gfx7_gfx8_gfx9_gfx10_gfx11_gfx12_gfx13<0x0f>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// GFX8 (VI), GFX9.
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
class SOP1_Real_vi<bits<8> op, SOP1_Pseudo ps> :
|
|
SOP1_Real<op, ps>,
|
|
Select_vi<ps.PseudoInstr>;
|
|
|
|
class SOP2_Real_vi<bits<7> op, SOP2_Pseudo ps> :
|
|
SOP2_Real32<op, ps>,
|
|
Select_vi<ps.PseudoInstr>;
|
|
|
|
class SOPK_Real_vi<bits<5> op, SOPK_Pseudo ps> :
|
|
SOPK_Real32<op, ps>,
|
|
Select_vi<ps.PseudoInstr>;
|
|
|
|
def S_MOV_B32_vi : SOP1_Real_vi <0x00, S_MOV_B32>;
|
|
def S_MOV_B64_vi : SOP1_Real_vi <0x01, S_MOV_B64>;
|
|
def S_CMOV_B32_vi : SOP1_Real_vi <0x02, S_CMOV_B32>;
|
|
def S_CMOV_B64_vi : SOP1_Real_vi <0x03, S_CMOV_B64>;
|
|
def S_NOT_B32_vi : SOP1_Real_vi <0x04, S_NOT_B32>;
|
|
def S_NOT_B64_vi : SOP1_Real_vi <0x05, S_NOT_B64>;
|
|
def S_WQM_B32_vi : SOP1_Real_vi <0x06, S_WQM_B32>;
|
|
def S_WQM_B64_vi : SOP1_Real_vi <0x07, S_WQM_B64>;
|
|
def S_BREV_B32_vi : SOP1_Real_vi <0x08, S_BREV_B32>;
|
|
def S_BREV_B64_vi : SOP1_Real_vi <0x09, S_BREV_B64>;
|
|
def S_BCNT0_I32_B32_vi : SOP1_Real_vi <0x0a, S_BCNT0_I32_B32>;
|
|
def S_BCNT0_I32_B64_vi : SOP1_Real_vi <0x0b, S_BCNT0_I32_B64>;
|
|
def S_BCNT1_I32_B32_vi : SOP1_Real_vi <0x0c, S_BCNT1_I32_B32>;
|
|
def S_BCNT1_I32_B64_vi : SOP1_Real_vi <0x0d, S_BCNT1_I32_B64>;
|
|
def S_FF0_I32_B32_vi : SOP1_Real_vi <0x0e, S_FF0_I32_B32>;
|
|
def S_FF0_I32_B64_vi : SOP1_Real_vi <0x0f, S_FF0_I32_B64>;
|
|
def S_FF1_I32_B32_vi : SOP1_Real_vi <0x10, S_FF1_I32_B32>;
|
|
def S_FF1_I32_B64_vi : SOP1_Real_vi <0x11, S_FF1_I32_B64>;
|
|
def S_FLBIT_I32_B32_vi : SOP1_Real_vi <0x12, S_FLBIT_I32_B32>;
|
|
def S_FLBIT_I32_B64_vi : SOP1_Real_vi <0x13, S_FLBIT_I32_B64>;
|
|
def S_FLBIT_I32_vi : SOP1_Real_vi <0x14, S_FLBIT_I32>;
|
|
def S_FLBIT_I32_I64_vi : SOP1_Real_vi <0x15, S_FLBIT_I32_I64>;
|
|
def S_SEXT_I32_I8_vi : SOP1_Real_vi <0x16, S_SEXT_I32_I8>;
|
|
def S_SEXT_I32_I16_vi : SOP1_Real_vi <0x17, S_SEXT_I32_I16>;
|
|
def S_BITSET0_B32_vi : SOP1_Real_vi <0x18, S_BITSET0_B32>;
|
|
def S_BITSET0_B64_vi : SOP1_Real_vi <0x19, S_BITSET0_B64>;
|
|
def S_BITSET1_B32_vi : SOP1_Real_vi <0x1a, S_BITSET1_B32>;
|
|
def S_BITSET1_B64_vi : SOP1_Real_vi <0x1b, S_BITSET1_B64>;
|
|
def S_GETPC_B64_vi : SOP1_Real_vi <0x1c, S_GETPC_B64>;
|
|
def S_SETPC_B64_vi : SOP1_Real_vi <0x1d, S_SETPC_B64>;
|
|
def S_SWAPPC_B64_vi : SOP1_Real_vi <0x1e, S_SWAPPC_B64>;
|
|
def S_RFE_B64_vi : SOP1_Real_vi <0x1f, S_RFE_B64>;
|
|
def S_AND_SAVEEXEC_B64_vi : SOP1_Real_vi <0x20, S_AND_SAVEEXEC_B64>;
|
|
def S_OR_SAVEEXEC_B64_vi : SOP1_Real_vi <0x21, S_OR_SAVEEXEC_B64>;
|
|
def S_XOR_SAVEEXEC_B64_vi : SOP1_Real_vi <0x22, S_XOR_SAVEEXEC_B64>;
|
|
def S_ANDN2_SAVEEXEC_B64_vi: SOP1_Real_vi <0x23, S_ANDN2_SAVEEXEC_B64>;
|
|
def S_ORN2_SAVEEXEC_B64_vi : SOP1_Real_vi <0x24, S_ORN2_SAVEEXEC_B64>;
|
|
def S_NAND_SAVEEXEC_B64_vi : SOP1_Real_vi <0x25, S_NAND_SAVEEXEC_B64>;
|
|
def S_NOR_SAVEEXEC_B64_vi : SOP1_Real_vi <0x26, S_NOR_SAVEEXEC_B64>;
|
|
def S_XNOR_SAVEEXEC_B64_vi : SOP1_Real_vi <0x27, S_XNOR_SAVEEXEC_B64>;
|
|
def S_QUADMASK_B32_vi : SOP1_Real_vi <0x28, S_QUADMASK_B32>;
|
|
def S_QUADMASK_B64_vi : SOP1_Real_vi <0x29, S_QUADMASK_B64>;
|
|
def S_MOVRELS_B32_vi : SOP1_Real_vi <0x2a, S_MOVRELS_B32>;
|
|
def S_MOVRELS_B64_vi : SOP1_Real_vi <0x2b, S_MOVRELS_B64>;
|
|
def S_MOVRELD_B32_vi : SOP1_Real_vi <0x2c, S_MOVRELD_B32>;
|
|
def S_MOVRELD_B64_vi : SOP1_Real_vi <0x2d, S_MOVRELD_B64>;
|
|
def S_CBRANCH_JOIN_vi : SOP1_Real_vi <0x2e, S_CBRANCH_JOIN>;
|
|
def S_ABS_I32_vi : SOP1_Real_vi <0x30, S_ABS_I32>;
|
|
def S_SET_GPR_IDX_IDX_vi : SOP1_Real_vi <0x32, S_SET_GPR_IDX_IDX>;
|
|
|
|
def S_ADD_U32_vi : SOP2_Real_vi <0x00, S_ADD_U32>;
|
|
def S_ADD_I32_vi : SOP2_Real_vi <0x02, S_ADD_I32>;
|
|
def S_SUB_U32_vi : SOP2_Real_vi <0x01, S_SUB_U32>;
|
|
def S_SUB_I32_vi : SOP2_Real_vi <0x03, S_SUB_I32>;
|
|
def S_ADDC_U32_vi : SOP2_Real_vi <0x04, S_ADDC_U32>;
|
|
def S_SUBB_U32_vi : SOP2_Real_vi <0x05, S_SUBB_U32>;
|
|
def S_MIN_I32_vi : SOP2_Real_vi <0x06, S_MIN_I32>;
|
|
def S_MIN_U32_vi : SOP2_Real_vi <0x07, S_MIN_U32>;
|
|
def S_MAX_I32_vi : SOP2_Real_vi <0x08, S_MAX_I32>;
|
|
def S_MAX_U32_vi : SOP2_Real_vi <0x09, S_MAX_U32>;
|
|
def S_CSELECT_B32_vi : SOP2_Real_vi <0x0a, S_CSELECT_B32>;
|
|
def S_CSELECT_B64_vi : SOP2_Real_vi <0x0b, S_CSELECT_B64>;
|
|
def S_AND_B32_vi : SOP2_Real_vi <0x0c, S_AND_B32>;
|
|
def S_AND_B64_vi : SOP2_Real_vi <0x0d, S_AND_B64>;
|
|
def S_OR_B32_vi : SOP2_Real_vi <0x0e, S_OR_B32>;
|
|
def S_OR_B64_vi : SOP2_Real_vi <0x0f, S_OR_B64>;
|
|
def S_XOR_B32_vi : SOP2_Real_vi <0x10, S_XOR_B32>;
|
|
def S_XOR_B64_vi : SOP2_Real_vi <0x11, S_XOR_B64>;
|
|
def S_ANDN2_B32_vi : SOP2_Real_vi <0x12, S_ANDN2_B32>;
|
|
def S_ANDN2_B64_vi : SOP2_Real_vi <0x13, S_ANDN2_B64>;
|
|
def S_ORN2_B32_vi : SOP2_Real_vi <0x14, S_ORN2_B32>;
|
|
def S_ORN2_B64_vi : SOP2_Real_vi <0x15, S_ORN2_B64>;
|
|
def S_NAND_B32_vi : SOP2_Real_vi <0x16, S_NAND_B32>;
|
|
def S_NAND_B64_vi : SOP2_Real_vi <0x17, S_NAND_B64>;
|
|
def S_NOR_B32_vi : SOP2_Real_vi <0x18, S_NOR_B32>;
|
|
def S_NOR_B64_vi : SOP2_Real_vi <0x19, S_NOR_B64>;
|
|
def S_XNOR_B32_vi : SOP2_Real_vi <0x1a, S_XNOR_B32>;
|
|
def S_XNOR_B64_vi : SOP2_Real_vi <0x1b, S_XNOR_B64>;
|
|
def S_LSHL_B32_vi : SOP2_Real_vi <0x1c, S_LSHL_B32>;
|
|
def S_LSHL_B64_vi : SOP2_Real_vi <0x1d, S_LSHL_B64>;
|
|
def S_LSHR_B32_vi : SOP2_Real_vi <0x1e, S_LSHR_B32>;
|
|
def S_LSHR_B64_vi : SOP2_Real_vi <0x1f, S_LSHR_B64>;
|
|
def S_ASHR_I32_vi : SOP2_Real_vi <0x20, S_ASHR_I32>;
|
|
def S_ASHR_I64_vi : SOP2_Real_vi <0x21, S_ASHR_I64>;
|
|
def S_BFM_B32_vi : SOP2_Real_vi <0x22, S_BFM_B32>;
|
|
def S_BFM_B64_vi : SOP2_Real_vi <0x23, S_BFM_B64>;
|
|
def S_MUL_I32_vi : SOP2_Real_vi <0x24, S_MUL_I32>;
|
|
def S_BFE_U32_vi : SOP2_Real_vi <0x25, S_BFE_U32>;
|
|
def S_BFE_I32_vi : SOP2_Real_vi <0x26, S_BFE_I32>;
|
|
def S_BFE_U64_vi : SOP2_Real_vi <0x27, S_BFE_U64>;
|
|
def S_BFE_I64_vi : SOP2_Real_vi <0x28, S_BFE_I64>;
|
|
def S_CBRANCH_G_FORK_vi : SOP2_Real_vi <0x29, S_CBRANCH_G_FORK>;
|
|
def S_ABSDIFF_I32_vi : SOP2_Real_vi <0x2a, S_ABSDIFF_I32>;
|
|
def S_PACK_LL_B32_B16_vi : SOP2_Real_vi <0x32, S_PACK_LL_B32_B16>;
|
|
def S_PACK_LH_B32_B16_vi : SOP2_Real_vi <0x33, S_PACK_LH_B32_B16>;
|
|
def S_PACK_HH_B32_B16_vi : SOP2_Real_vi <0x34, S_PACK_HH_B32_B16>;
|
|
def S_RFE_RESTORE_B64_vi : SOP2_Real_vi <0x2b, S_RFE_RESTORE_B64>;
|
|
|
|
def S_MOVK_I32_vi : SOPK_Real_vi <0x00, S_MOVK_I32>;
|
|
def S_CMOVK_I32_vi : SOPK_Real_vi <0x01, S_CMOVK_I32>;
|
|
def S_CMPK_EQ_I32_vi : SOPK_Real_vi <0x02, S_CMPK_EQ_I32>;
|
|
def S_CMPK_LG_I32_vi : SOPK_Real_vi <0x03, S_CMPK_LG_I32>;
|
|
def S_CMPK_GT_I32_vi : SOPK_Real_vi <0x04, S_CMPK_GT_I32>;
|
|
def S_CMPK_GE_I32_vi : SOPK_Real_vi <0x05, S_CMPK_GE_I32>;
|
|
def S_CMPK_LT_I32_vi : SOPK_Real_vi <0x06, S_CMPK_LT_I32>;
|
|
def S_CMPK_LE_I32_vi : SOPK_Real_vi <0x07, S_CMPK_LE_I32>;
|
|
def S_CMPK_EQ_U32_vi : SOPK_Real_vi <0x08, S_CMPK_EQ_U32>;
|
|
def S_CMPK_LG_U32_vi : SOPK_Real_vi <0x09, S_CMPK_LG_U32>;
|
|
def S_CMPK_GT_U32_vi : SOPK_Real_vi <0x0A, S_CMPK_GT_U32>;
|
|
def S_CMPK_GE_U32_vi : SOPK_Real_vi <0x0B, S_CMPK_GE_U32>;
|
|
def S_CMPK_LT_U32_vi : SOPK_Real_vi <0x0C, S_CMPK_LT_U32>;
|
|
def S_CMPK_LE_U32_vi : SOPK_Real_vi <0x0D, S_CMPK_LE_U32>;
|
|
def S_ADDK_I32_vi : SOPK_Real_vi <0x0E, S_ADDK_I32>;
|
|
def S_MULK_I32_vi : SOPK_Real_vi <0x0F, S_MULK_I32>;
|
|
def S_CBRANCH_I_FORK_vi : SOPK_Real_vi <0x10, S_CBRANCH_I_FORK>;
|
|
def S_GETREG_B32_vi : SOPK_Real_vi <0x11, S_GETREG_B32>;
|
|
def S_SETREG_B32_vi : SOPK_Real_vi <0x12, S_SETREG_B32>;
|
|
//def S_GETREG_REGRD_B32_vi : SOPK_Real_vi <0x13, S_GETREG_REGRD_B32>; // see pseudo for comments
|
|
def S_SETREG_IMM32_B32_vi : SOPK_Real64<0x14, S_SETREG_IMM32_B32>,
|
|
Select_vi<S_SETREG_IMM32_B32.PseudoInstr>;
|
|
|
|
def S_CALL_B64_vi : SOPK_Real_vi <0x15, S_CALL_B64>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOP1 - GFX9.
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
def S_ANDN1_SAVEEXEC_B64_vi : SOP1_Real_vi<0x33, S_ANDN1_SAVEEXEC_B64>;
|
|
def S_ORN1_SAVEEXEC_B64_vi : SOP1_Real_vi<0x34, S_ORN1_SAVEEXEC_B64>;
|
|
def S_ANDN1_WREXEC_B64_vi : SOP1_Real_vi<0x35, S_ANDN1_WREXEC_B64>;
|
|
def S_ANDN2_WREXEC_B64_vi : SOP1_Real_vi<0x36, S_ANDN2_WREXEC_B64>;
|
|
def S_BITREPLICATE_B64_B32_vi : SOP1_Real_vi<0x37, S_BITREPLICATE_B64_B32>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// SOP2 - GFX9.
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
def S_LSHL1_ADD_U32_vi : SOP2_Real_vi<0x2e, S_LSHL1_ADD_U32>;
|
|
def S_LSHL2_ADD_U32_vi : SOP2_Real_vi<0x2f, S_LSHL2_ADD_U32>;
|
|
def S_LSHL3_ADD_U32_vi : SOP2_Real_vi<0x30, S_LSHL3_ADD_U32>;
|
|
def S_LSHL4_ADD_U32_vi : SOP2_Real_vi<0x31, S_LSHL4_ADD_U32>;
|
|
def S_MUL_HI_U32_vi : SOP2_Real_vi<0x2c, S_MUL_HI_U32>;
|
|
def S_MUL_HI_I32_vi : SOP2_Real_vi<0x2d, S_MUL_HI_I32>;
|