Skip to content

JIT ARM64-SVE: Add simple bitwise ops #101762

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 5 commits into from
May 3, 2024
Merged

JIT ARM64-SVE: Add simple bitwise ops #101762

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
c442956
JIT ARM64-SVE: Add simple bitwise ops
a74nh May 1, 2024
f312257
Fix fadda
a74nh May 2, 2024
f26f7ce
Fix unpkh/fexpa/frecpe
a74nh May 2, 2024
7a23e95
Reorder System.Runtime.Intrinsics.cs
a74nh May 2, 2024
8a92556
Fix API head comments
a74nh May 2, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
46 changes: 23 additions & 23 deletions src/coreclr/jit/codegenarm64test.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4802,11 +4802,11 @@ void CodeGen::genArm64EmitterUnitTestsSve()
INS_OPTS_SCALABLE_D); // CLASTB <Zdn>.<T>, <Pg>, <Zdn>.<T>, <Zm>.<T>

// IF_SVE_CN_3A
theEmitter->emitIns_R_R_R(INS_sve_clasta, EA_2BYTE, REG_V12, REG_P1, REG_V15, INS_OPTS_SCALABLE_H,
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Same changes as for AddAcross in previous PR - the size arg is not used, as the sizes are dependant on opts.

theEmitter->emitIns_R_R_R(INS_sve_clasta, EA_SCALABLE, REG_V12, REG_P1, REG_V15, INS_OPTS_SCALABLE_H,
INS_SCALABLE_OPTS_WITH_SIMD_SCALAR); // CLASTA <V><dn>, <Pg>, <V><dn>, <Zm>.<T>
theEmitter->emitIns_R_R_R(INS_sve_clastb, EA_4BYTE, REG_V13, REG_P2, REG_V16, INS_OPTS_SCALABLE_S,
theEmitter->emitIns_R_R_R(INS_sve_clastb, EA_SCALABLE, REG_V13, REG_P2, REG_V16, INS_OPTS_SCALABLE_S,
INS_SCALABLE_OPTS_WITH_SIMD_SCALAR); // CLASTB <V><dn>, <Pg>, <V><dn>, <Zm>.<T>
theEmitter->emitIns_R_R_R(INS_sve_clastb, EA_8BYTE, REG_V14, REG_P0, REG_V17, INS_OPTS_SCALABLE_D,
theEmitter->emitIns_R_R_R(INS_sve_clastb, EA_SCALABLE, REG_V14, REG_P0, REG_V17, INS_OPTS_SCALABLE_D,
INS_SCALABLE_OPTS_WITH_SIMD_SCALAR); // CLASTB <V><dn>, <Pg>, <V><dn>, <Zm>.<T>

// IF_SVE_CO_3A
Expand Down Expand Up @@ -5116,11 +5116,11 @@ void CodeGen::genArm64EmitterUnitTestsSve()
INS_OPTS_SCALABLE_H); // FMINP <Zdn>.<T>, <Pg>/M, <Zdn>.<T>, <Zm>.<T>

// IF_SVE_HJ_3A
theEmitter->emitIns_R_R_R(INS_sve_fadda, EA_2BYTE, REG_V21, REG_P6, REG_V14,
theEmitter->emitIns_R_R_R(INS_sve_fadda, EA_SCALABLE, REG_V21, REG_P6, REG_V14,
INS_OPTS_SCALABLE_H); // FADDA <V><dn>, <Pg>, <V><dn>, <Zm>.<T>
theEmitter->emitIns_R_R_R(INS_sve_fadda, EA_4BYTE, REG_V22, REG_P5, REG_V13,
theEmitter->emitIns_R_R_R(INS_sve_fadda, EA_SCALABLE, REG_V22, REG_P5, REG_V13,
INS_OPTS_SCALABLE_S); // FADDA <V><dn>, <Pg>, <V><dn>, <Zm>.<T>
theEmitter->emitIns_R_R_R(INS_sve_fadda, EA_8BYTE, REG_V23, REG_P4, REG_V12,
theEmitter->emitIns_R_R_R(INS_sve_fadda, EA_SCALABLE, REG_V23, REG_P4, REG_V12,
INS_OPTS_SCALABLE_D); // FADDA <V><dn>, <Pg>, <V><dn>, <Zm>.<T>

// IF_SVE_HL_3A
Expand Down Expand Up @@ -5292,13 +5292,13 @@ void CodeGen::genArm64EmitterUnitTestsSve()
INS_OPTS_SCALABLE_H); // FNMLS <Zda>.<T>, <Pg>/M, <Zn>.<T>, <Zm>.<T>

// IF_SVE_AF_3A
theEmitter->emitIns_R_R_R(INS_sve_andv, EA_1BYTE, REG_V0, REG_P0, REG_V0,
theEmitter->emitIns_R_R_R(INS_sve_andv, EA_SCALABLE, REG_V0, REG_P0, REG_V0,
INS_OPTS_SCALABLE_B); // ANDV <V><d>, <Pg>, <Zn>.<T>
theEmitter->emitIns_R_R_R(INS_sve_eorv, EA_2BYTE, REG_V1, REG_P1, REG_V1,
theEmitter->emitIns_R_R_R(INS_sve_eorv, EA_SCALABLE, REG_V1, REG_P1, REG_V1,
INS_OPTS_SCALABLE_H); // EORV <V><d>, <Pg>, <Zn>.<T>
theEmitter->emitIns_R_R_R(INS_sve_orv, EA_4BYTE, REG_V2, REG_P2, REG_V2,
theEmitter->emitIns_R_R_R(INS_sve_orv, EA_SCALABLE, REG_V2, REG_P2, REG_V2,
INS_OPTS_SCALABLE_S); // ORV <V><d>, <Pg>, <Zn>.<T>
theEmitter->emitIns_R_R_R(INS_sve_orv, EA_8BYTE, REG_V3, REG_P3, REG_V3,
theEmitter->emitIns_R_R_R(INS_sve_orv, EA_SCALABLE, REG_V3, REG_P3, REG_V3,
INS_OPTS_SCALABLE_D); // ORV <V><d>, <Pg>, <Zn>.<T>

// IF_SVE_AG_3A
Expand Down Expand Up @@ -5328,13 +5328,13 @@ void CodeGen::genArm64EmitterUnitTestsSve()
#endif // ALL_ARM64_EMITTER_UNIT_TESTS_SVE_UNSUPPORTED

// IF_SVE_AK_3A
theEmitter->emitIns_R_R_R(INS_sve_smaxv, EA_8BYTE, REG_V15, REG_P7, REG_V4,
theEmitter->emitIns_R_R_R(INS_sve_smaxv, EA_SCALABLE, REG_V15, REG_P7, REG_V4,
INS_OPTS_SCALABLE_D); // SMAXV <V><d>, <Pg>, <Zn>.<T>
theEmitter->emitIns_R_R_R(INS_sve_sminv, EA_4BYTE, REG_V16, REG_P6, REG_V14,
theEmitter->emitIns_R_R_R(INS_sve_sminv, EA_SCALABLE, REG_V16, REG_P6, REG_V14,
INS_OPTS_SCALABLE_S); // SMINV <V><d>, <Pg>, <Zn>.<T>
theEmitter->emitIns_R_R_R(INS_sve_umaxv, EA_2BYTE, REG_V17, REG_P5, REG_V24,
theEmitter->emitIns_R_R_R(INS_sve_umaxv, EA_SCALABLE, REG_V17, REG_P5, REG_V24,
INS_OPTS_SCALABLE_H); // UMAXV <V><d>, <Pg>, <Zn>.<T>
theEmitter->emitIns_R_R_R(INS_sve_uminv, EA_1BYTE, REG_V18, REG_P4, REG_V31,
theEmitter->emitIns_R_R_R(INS_sve_uminv, EA_SCALABLE, REG_V18, REG_P4, REG_V31,
INS_OPTS_SCALABLE_B); // UMINV <V><d>, <Pg>, <Zn>.<T>

// IF_SVE_AL_3A
Expand Down Expand Up @@ -6222,13 +6222,13 @@ void CodeGen::genArm64EmitterUnitTestsSve()
INS_OPTS_SCALABLE_D); // COMPACT <Zd>.<T>, <Pg>, <Zn>.<T>

// IF_SVE_CP_3A
theEmitter->emitIns_R_R_R(INS_sve_cpy, EA_1BYTE, REG_V14, REG_P1, REG_V11, INS_OPTS_SCALABLE_B,
theEmitter->emitIns_R_R_R(INS_sve_cpy, EA_SCALABLE, REG_V14, REG_P1, REG_V11, INS_OPTS_SCALABLE_B,
INS_SCALABLE_OPTS_WITH_SIMD_SCALAR); // CPY <Zd>.<T>, <Pg>/M, <V><n>
theEmitter->emitIns_R_R_R(INS_sve_cpy, EA_4BYTE, REG_V13, REG_P2, REG_V10, INS_OPTS_SCALABLE_S,
theEmitter->emitIns_R_R_R(INS_sve_cpy, EA_SCALABLE, REG_V13, REG_P2, REG_V10, INS_OPTS_SCALABLE_S,
INS_SCALABLE_OPTS_WITH_SIMD_SCALAR); // CPY <Zd>.<T>, <Pg>/M, <V><n>
theEmitter->emitIns_R_R_R(INS_sve_mov, EA_2BYTE, REG_V12, REG_P3, REG_V9, INS_OPTS_SCALABLE_H,
theEmitter->emitIns_R_R_R(INS_sve_mov, EA_SCALABLE, REG_V12, REG_P3, REG_V9, INS_OPTS_SCALABLE_H,
INS_SCALABLE_OPTS_WITH_SIMD_SCALAR); // MOV <Zd>.<T>, <Pg>/M, <V><n>
theEmitter->emitIns_R_R_R(INS_sve_mov, EA_8BYTE, REG_V11, REG_P4, REG_V8, INS_OPTS_SCALABLE_D,
theEmitter->emitIns_R_R_R(INS_sve_mov, EA_SCALABLE, REG_V11, REG_P4, REG_V8, INS_OPTS_SCALABLE_D,
INS_SCALABLE_OPTS_WITH_SIMD_SCALAR); // MOV <Zd>.<T>, <Pg>/M, <V><n>

// IF_SVE_CQ_3A
Expand All @@ -6243,13 +6243,13 @@ void CodeGen::genArm64EmitterUnitTestsSve()
INS_OPTS_SCALABLE_B); // MOV <Zd>.<T>, <Pg>/M, <R><n|SP>

// IF_SVE_CR_3A
theEmitter->emitIns_R_R_R(INS_sve_lasta, EA_1BYTE, REG_V6, REG_P1, REG_V27, INS_OPTS_SCALABLE_B,
theEmitter->emitIns_R_R_R(INS_sve_lasta, EA_SCALABLE, REG_V6, REG_P1, REG_V27, INS_OPTS_SCALABLE_B,
INS_SCALABLE_OPTS_WITH_SIMD_SCALAR); // LASTA <V><d>, <Pg>, <Zn>.<T>
theEmitter->emitIns_R_R_R(INS_sve_lasta, EA_2BYTE, REG_V5, REG_P2, REG_V26, INS_OPTS_SCALABLE_H,
theEmitter->emitIns_R_R_R(INS_sve_lasta, EA_SCALABLE, REG_V5, REG_P2, REG_V26, INS_OPTS_SCALABLE_H,
INS_SCALABLE_OPTS_WITH_SIMD_SCALAR); // LASTA <V><d>, <Pg>, <Zn>.<T>
theEmitter->emitIns_R_R_R(INS_sve_lastb, EA_4BYTE, REG_V4, REG_P3, REG_V25, INS_OPTS_SCALABLE_S,
theEmitter->emitIns_R_R_R(INS_sve_lastb, EA_SCALABLE, REG_V4, REG_P3, REG_V25, INS_OPTS_SCALABLE_S,
INS_SCALABLE_OPTS_WITH_SIMD_SCALAR); // LASTB <V><d>, <Pg>, <Zn>.<T>
theEmitter->emitIns_R_R_R(INS_sve_lastb, EA_8BYTE, REG_V3, REG_P4, REG_V24, INS_OPTS_SCALABLE_D,
theEmitter->emitIns_R_R_R(INS_sve_lastb, EA_SCALABLE, REG_V3, REG_P4, REG_V24, INS_OPTS_SCALABLE_D,
INS_SCALABLE_OPTS_WITH_SIMD_SCALAR); // LASTB <V><d>, <Pg>, <Zn>.<T>

// IF_SVE_CS_3A
Expand Down Expand Up @@ -8909,7 +8909,7 @@ void CodeGen::genArm64EmitterUnitTestsSve()
// SUNPKLO <Zd>.<T>, <Zn>.<Tb>
theEmitter->emitIns_R_R(INS_sve_sunpklo, EA_SCALABLE, REG_V1, REG_V5, INS_OPTS_SCALABLE_S);
// UUNPKHI <Zd>.<T>, <Zn>.<Tb>
theEmitter->emitIns_R_R(INS_sve_uunpkhi, EA_SCALABLE, REG_V5, REG_V1, INS_OPTS_SCALABLE_D);
theEmitter->emitIns_R_R(INS_sve_uunpkhi, EA_SCALABLE, REG_V5, REG_V1, INS_OPTS_SCALABLE_B);
// UUNPKLO <Zd>.<T>, <Zn>.<Tb>
theEmitter->emitIns_R_R(INS_sve_uunpklo, EA_SCALABLE, REG_V8, REG_V6, INS_OPTS_SCALABLE_S);

Expand Down
19 changes: 14 additions & 5 deletions src/coreclr/jit/emitarm64sve.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2919,7 +2919,10 @@ void emitter::emitInsSve_R_R_R(instruction ins,

if (sopt == INS_SCALABLE_OPTS_UNPREDICATED)
{
assert(opt == INS_OPTS_SCALABLE_D);
// The instruction only has a .D variant. However, this doesn't matter as
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Doing this prevents adding special cases in hwinstrinccodegen.

// it operates on bits not lanes. Effectively this means all standard opt
// sizes are supported.
assert(insOptsScalableStandard(opt));
assert(isVectorRegister(reg2)); // nnnnn
fmt = IF_SVE_AU_3A;
}
Expand Down Expand Up @@ -3728,7 +3731,7 @@ void emitter::emitInsSve_R_R_R(instruction ins,
else if (sopt == INS_SCALABLE_OPTS_WITH_SIMD_SCALAR)
{
assert(isFloatReg(reg1));
assert(isValidVectorElemsize(size));
assert(isScalableVectorSize(size));
fmt = IF_SVE_CN_3A;
}
else
Expand All @@ -3748,7 +3751,7 @@ void emitter::emitInsSve_R_R_R(instruction ins,
if (sopt == INS_SCALABLE_OPTS_UNPREDICATED)
{
assert(ins == INS_sve_mov);
assert(opt == INS_OPTS_SCALABLE_D);
assert(insOptsScalableStandard(opt));
assert(isVectorRegister(reg1)); // ddddd
assert(isVectorRegister(reg2)); // nnnnn
assert(isVectorRegister(reg3)); // mmmmm
Expand Down Expand Up @@ -11820,6 +11823,7 @@ BYTE* emitter::emitOutput_InstrSve(BYTE* dst, instrDesc* id)
code |= insEncodeSveElemsize(optGetSveElemsize((insOpts)(id->idInsOpt() + 1))); // xx
dst += emitOutput_Instr(dst, code);
break;

case IF_SVE_BF_2A: // ........xx.xxiii ......nnnnnddddd -- SVE bitwise shift by immediate (unpredicated)
case IF_SVE_FT_2A: // ........xx.xxiii ......nnnnnddddd -- SVE2 bitwise shift and insert
case IF_SVE_FU_2A: // ........xx.xxiii ......nnnnnddddd -- SVE2 bitwise shift right and accumulate
Expand Down Expand Up @@ -12622,7 +12626,7 @@ void emitter::emitInsSveSanityCheck(instrDesc* id)
assert(isVectorRegister(id->idReg1())); // ddddd
assert(isLowPredicateRegister(id->idReg2())); // ggg
assert(isVectorRegister(id->idReg3())); // mmmmm
assert(isValidVectorElemsize(id->idOpSize()));
assert(isScalableVectorSize(id->idOpSize()));
break;

// Scalable to FP SIMD scalar.
Expand Down Expand Up @@ -14393,8 +14397,13 @@ void emitter::emitInsSveSanityCheck(instrDesc* id)
break;

case IF_SVE_BJ_2A: // ........xx...... ......nnnnnddddd -- SVE floating-point exponential accelerator
case IF_SVE_CH_2A: // ........xx...... ......nnnnnddddd -- SVE unpack vector elements
case IF_SVE_HF_2A: // ........xx...... ......nnnnnddddd -- SVE floating-point reciprocal estimate (unpredicated)
assert(insOptsScalableAtLeastHalf(id->idInsOpt()));
assert(isVectorRegister(id->idReg1()));
assert(isVectorRegister(id->idReg2()));
break;

case IF_SVE_CH_2A: // ........xx...... ......nnnnnddddd -- SVE unpack vector elements
assert(insOptsScalableWide(id->idInsOpt()));
assert(isVectorRegister(id->idReg1()));
assert(isVectorRegister(id->idReg2()));
Expand Down
Loading
Loading