IR: Implement VectorNarrow
This commit is contained in:
parent
1423584f9f
commit
132c783320
4 changed files with 58 additions and 0 deletions
|
@ -414,6 +414,47 @@ void EmitX64::EmitVectorInterleaveLower64(EmitContext& ctx, IR::Inst* inst) {
|
||||||
EmitVectorInterleaveLower(code, ctx, inst, 64);
|
EmitVectorInterleaveLower(code, ctx, inst, 64);
|
||||||
}
|
}
|
||||||
|
|
||||||
|
void EmitX64::EmitVectorNarrow16(EmitContext& ctx, IR::Inst* inst) {
|
||||||
|
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
|
||||||
|
Xbyak::Xmm a = ctx.reg_alloc.UseScratchXmm(args[0]);
|
||||||
|
Xbyak::Xmm zeros = ctx.reg_alloc.ScratchXmm();
|
||||||
|
|
||||||
|
// TODO: AVX512F implementation
|
||||||
|
|
||||||
|
code.pxor(zeros, zeros);
|
||||||
|
code.pand(a, code.MConst(0x00FF00FF00FF00FF, 0x00FF00FF00FF00FF));
|
||||||
|
code.packuswb(a, zeros);
|
||||||
|
|
||||||
|
ctx.reg_alloc.DefineValue(inst, a);
|
||||||
|
}
|
||||||
|
|
||||||
|
void EmitX64::EmitVectorNarrow32(EmitContext& ctx, IR::Inst* inst) {
|
||||||
|
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
|
||||||
|
Xbyak::Xmm a = ctx.reg_alloc.UseScratchXmm(args[0]);
|
||||||
|
Xbyak::Xmm zeros = ctx.reg_alloc.ScratchXmm();
|
||||||
|
|
||||||
|
// TODO: AVX512F implementation
|
||||||
|
|
||||||
|
code.pxor(zeros, zeros);
|
||||||
|
code.pand(a, code.MConst(0x0000FFFF0000FFFF, 0x0000FFFF0000FFFF));
|
||||||
|
code.packusdw(a, zeros);
|
||||||
|
|
||||||
|
ctx.reg_alloc.DefineValue(inst, a);
|
||||||
|
}
|
||||||
|
|
||||||
|
void EmitX64::EmitVectorNarrow64(EmitContext& ctx, IR::Inst* inst) {
|
||||||
|
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
|
||||||
|
Xbyak::Xmm a = ctx.reg_alloc.UseScratchXmm(args[0]);
|
||||||
|
Xbyak::Xmm zeros = ctx.reg_alloc.ScratchXmm();
|
||||||
|
|
||||||
|
// TODO: AVX512F implementation
|
||||||
|
|
||||||
|
code.pxor(zeros, zeros);
|
||||||
|
code.shufps(a, zeros, 0b00001000);
|
||||||
|
|
||||||
|
ctx.reg_alloc.DefineValue(inst, a);
|
||||||
|
}
|
||||||
|
|
||||||
void EmitX64::EmitVectorNot(EmitContext& ctx, IR::Inst* inst) {
|
void EmitX64::EmitVectorNot(EmitContext& ctx, IR::Inst* inst) {
|
||||||
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
|
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
|
||||||
|
|
||||||
|
|
|
@ -898,6 +898,19 @@ U128 IREmitter::VectorLogicalShiftRight(size_t esize, const U128& a, u8 shift_am
|
||||||
return {};
|
return {};
|
||||||
}
|
}
|
||||||
|
|
||||||
|
U128 IREmitter::VectorNarrow(size_t original_esize, const U128& a) {
|
||||||
|
switch (original_esize) {
|
||||||
|
case 16:
|
||||||
|
return Inst<U128>(Opcode::VectorNarrow16, a);
|
||||||
|
case 32:
|
||||||
|
return Inst<U128>(Opcode::VectorNarrow32, a);
|
||||||
|
case 64:
|
||||||
|
return Inst<U128>(Opcode::VectorNarrow64, a);
|
||||||
|
}
|
||||||
|
UNREACHABLE();
|
||||||
|
return {};
|
||||||
|
}
|
||||||
|
|
||||||
U128 IREmitter::VectorNot(const U128& a) {
|
U128 IREmitter::VectorNot(const U128& a) {
|
||||||
return Inst<U128>(Opcode::VectorNot, a);
|
return Inst<U128>(Opcode::VectorNot, a);
|
||||||
}
|
}
|
||||||
|
|
|
@ -216,6 +216,7 @@ public:
|
||||||
U128 VectorInterleaveLower(size_t esize, const U128& a, const U128& b);
|
U128 VectorInterleaveLower(size_t esize, const U128& a, const U128& b);
|
||||||
U128 VectorLogicalShiftLeft(size_t esize, const U128& a, u8 shift_amount);
|
U128 VectorLogicalShiftLeft(size_t esize, const U128& a, u8 shift_amount);
|
||||||
U128 VectorLogicalShiftRight(size_t esize, const U128& a, u8 shift_amount);
|
U128 VectorLogicalShiftRight(size_t esize, const U128& a, u8 shift_amount);
|
||||||
|
U128 VectorNarrow(size_t original_esize, const U128& a);
|
||||||
U128 VectorNot(const U128& a);
|
U128 VectorNot(const U128& a);
|
||||||
U128 VectorOr(const U128& a, const U128& b);
|
U128 VectorOr(const U128& a, const U128& b);
|
||||||
U128 VectorPairedAdd(size_t esize, const U128& a, const U128& b);
|
U128 VectorPairedAdd(size_t esize, const U128& a, const U128& b);
|
||||||
|
|
|
@ -228,6 +228,9 @@ OPCODE(VectorLogicalShiftRight8, T::U128, T::U128, T::U8
|
||||||
OPCODE(VectorLogicalShiftRight16, T::U128, T::U128, T::U8 )
|
OPCODE(VectorLogicalShiftRight16, T::U128, T::U128, T::U8 )
|
||||||
OPCODE(VectorLogicalShiftRight32, T::U128, T::U128, T::U8 )
|
OPCODE(VectorLogicalShiftRight32, T::U128, T::U128, T::U8 )
|
||||||
OPCODE(VectorLogicalShiftRight64, T::U128, T::U128, T::U8 )
|
OPCODE(VectorLogicalShiftRight64, T::U128, T::U128, T::U8 )
|
||||||
|
OPCODE(VectorNarrow16, T::U128, T::U128 )
|
||||||
|
OPCODE(VectorNarrow32, T::U128, T::U128 )
|
||||||
|
OPCODE(VectorNarrow64, T::U128, T::U128 )
|
||||||
OPCODE(VectorNot, T::U128, T::U128 )
|
OPCODE(VectorNot, T::U128, T::U128 )
|
||||||
OPCODE(VectorOr, T::U128, T::U128, T::U128 )
|
OPCODE(VectorOr, T::U128, T::U128, T::U128 )
|
||||||
OPCODE(VectorPairedAddLower8, T::U128, T::U128, T::U128 )
|
OPCODE(VectorPairedAddLower8, T::U128, T::U128, T::U128 )
|
||||||
|
|
Loading…
Reference in a new issue