IR: Add opcodes for interleaving upper-order bytes/halfwords/words/doublewords
I should have added this when I introduced the functions for interleaving low-order equivalents for consistency in the interface.
This commit is contained in:
parent
94f0fba16b
commit
701f43d61e
4 changed files with 60 additions and 0 deletions
|
@ -552,6 +552,46 @@ void EmitX64::EmitVectorInterleaveLower64(EmitContext& ctx, IR::Inst* inst) {
|
|||
EmitVectorInterleaveLower(code, ctx, inst, 64);
|
||||
}
|
||||
|
||||
static void EmitVectorInterleaveUpper(BlockOfCode& code, EmitContext& ctx, IR::Inst* inst, int size) {
|
||||
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
|
||||
|
||||
const Xbyak::Xmm a = ctx.reg_alloc.UseScratchXmm(args[0]);
|
||||
const Xbyak::Xmm b = ctx.reg_alloc.UseXmm(args[1]);
|
||||
|
||||
switch (size) {
|
||||
case 8:
|
||||
code.punpckhbw(a, b);
|
||||
break;
|
||||
case 16:
|
||||
code.punpckhwd(a, b);
|
||||
break;
|
||||
case 32:
|
||||
code.punpckhdq(a, b);
|
||||
break;
|
||||
case 64:
|
||||
code.punpckhqdq(a, b);
|
||||
break;
|
||||
}
|
||||
|
||||
ctx.reg_alloc.DefineValue(inst, a);
|
||||
}
|
||||
|
||||
void EmitX64::EmitVectorInterleaveUpper8(EmitContext& ctx, IR::Inst* inst) {
|
||||
EmitVectorInterleaveUpper(code, ctx, inst, 8);
|
||||
}
|
||||
|
||||
void EmitX64::EmitVectorInterleaveUpper16(EmitContext& ctx, IR::Inst* inst) {
|
||||
EmitVectorInterleaveUpper(code, ctx, inst, 16);
|
||||
}
|
||||
|
||||
void EmitX64::EmitVectorInterleaveUpper32(EmitContext& ctx, IR::Inst* inst) {
|
||||
EmitVectorInterleaveUpper(code, ctx, inst, 32);
|
||||
}
|
||||
|
||||
void EmitX64::EmitVectorInterleaveUpper64(EmitContext& ctx, IR::Inst* inst) {
|
||||
EmitVectorInterleaveUpper(code, ctx, inst, 64);
|
||||
}
|
||||
|
||||
void EmitX64::EmitVectorLogicalShiftLeft8(EmitContext& ctx, IR::Inst* inst) {
|
||||
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
|
||||
|
||||
|
|
|
@ -914,6 +914,21 @@ U128 IREmitter::VectorInterleaveLower(size_t esize, const U128& a, const U128& b
|
|||
return {};
|
||||
}
|
||||
|
||||
U128 IREmitter::VectorInterleaveUpper(size_t esize, const U128& a, const U128& b) {
|
||||
switch (esize) {
|
||||
case 8:
|
||||
return Inst<U128>(Opcode::VectorInterleaveUpper8, a, b);
|
||||
case 16:
|
||||
return Inst<U128>(Opcode::VectorInterleaveUpper16, a, b);
|
||||
case 32:
|
||||
return Inst<U128>(Opcode::VectorInterleaveUpper32, a, b);
|
||||
case 64:
|
||||
return Inst<U128>(Opcode::VectorInterleaveUpper64, a, b);
|
||||
}
|
||||
UNREACHABLE();
|
||||
return {};
|
||||
}
|
||||
|
||||
U128 IREmitter::VectorLessEqualSigned(size_t esize, const U128& a, const U128& b) {
|
||||
return VectorNot(VectorGreaterSigned(esize, a, b));
|
||||
}
|
||||
|
|
|
@ -220,6 +220,7 @@ public:
|
|||
U128 VectorGreaterSigned(size_t esize, const U128& a, const U128& b);
|
||||
U128 VectorGreaterUnsigned(size_t esize, const U128& a, const U128& b);
|
||||
U128 VectorInterleaveLower(size_t esize, const U128& a, const U128& b);
|
||||
U128 VectorInterleaveUpper(size_t esize, const U128& a, const U128& b);
|
||||
U128 VectorLessEqualSigned(size_t esize, const U128& a, const U128& b);
|
||||
U128 VectorLessEqualUnsigned(size_t esize, const U128& a, const U128& b);
|
||||
U128 VectorLessSigned(size_t esize, const U128& a, const U128& b);
|
||||
|
|
|
@ -242,6 +242,10 @@ OPCODE(VectorInterleaveLower8, T::U128, T::U128, T::U128
|
|||
OPCODE(VectorInterleaveLower16, T::U128, T::U128, T::U128 )
|
||||
OPCODE(VectorInterleaveLower32, T::U128, T::U128, T::U128 )
|
||||
OPCODE(VectorInterleaveLower64, T::U128, T::U128, T::U128 )
|
||||
OPCODE(VectorInterleaveUpper8, T::U128, T::U128, T::U128 )
|
||||
OPCODE(VectorInterleaveUpper16, T::U128, T::U128, T::U128 )
|
||||
OPCODE(VectorInterleaveUpper32, T::U128, T::U128, T::U128 )
|
||||
OPCODE(VectorInterleaveUpper64, T::U128, T::U128, T::U128 )
|
||||
OPCODE(VectorLogicalShiftLeft8, T::U128, T::U128, T::U8 )
|
||||
OPCODE(VectorLogicalShiftLeft16, T::U128, T::U128, T::U8 )
|
||||
OPCODE(VectorLogicalShiftLeft32, T::U128, T::U128, T::U8 )
|
||||
|
|
Loading…
Reference in a new issue