mirrors/dynarmic
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

IR: Implement VectorTranspose

Browse source
This commit is contained in:
MerryMage 2020-06-21 11:10:14 +01:00
parent 9cc11681dc
commit 7d1e103ff5
4 changed files with 82 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

63
src/backend/x64/emit_x64_vector.cpp
View file

@ -4326,6 +4326,69 @@ void EmitX64::EmitVectorTableLookup128(EmitContext& ctx, IR::Inst* inst) {
ctx.reg_alloc.DefineValue(inst, result);
}
void EmitX64::EmitVectorTranspose8(EmitContext& ctx, IR::Inst* inst) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm lower = ctx.reg_alloc.UseScratchXmm(args[0]);
const Xbyak::Xmm upper = ctx.reg_alloc.UseScratchXmm(args[1]);
const bool part = args[2].GetImmediateU1();
if (!part) {
code.pand(lower, code.MConst(xword, 0x00FF00FF00FF00FF, 0x00FF00FF00FF00FF));
code.psllw(upper, 8);
} else {
code.psrlw(lower, 8);
code.pand(upper, code.MConst(xword, 0xFF00FF00FF00FF00, 0xFF00FF00FF00FF00));
}
code.por(lower, upper);
ctx.reg_alloc.DefineValue(inst, lower);
}
void EmitX64::EmitVectorTranspose16(EmitContext& ctx, IR::Inst* inst) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm lower = ctx.reg_alloc.UseScratchXmm(args[0]);
const Xbyak::Xmm upper = ctx.reg_alloc.UseScratchXmm(args[1]);
const bool part = args[2].GetImmediateU1();
if (!part) {
code.pand(lower, code.MConst(xword, 0x0000FFFF0000FFFF, 0x0000FFFF0000FFFF));
code.pslld(upper, 16);
} else {
code.psrld(lower, 16);
code.pand(upper, code.MConst(xword, 0xFFFF0000FFFF0000, 0xFFFF0000FFFF0000));
}
code.por(lower, upper);
ctx.reg_alloc.DefineValue(inst, lower);
}
void EmitX64::EmitVectorTranspose32(EmitContext& ctx, IR::Inst* inst) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm lower = ctx.reg_alloc.UseScratchXmm(args[0]);
const Xbyak::Xmm upper = ctx.reg_alloc.UseXmm(args[1]);
const bool part = args[2].GetImmediateU1();
code.shufps(lower, upper, !part ? 0b10001000 : 0b11011101);
code.pshufd(lower, lower, 0b11011000);
ctx.reg_alloc.DefineValue(inst, lower);
}
void EmitX64::EmitVectorTranspose64(EmitContext& ctx, IR::Inst* inst) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm lower = ctx.reg_alloc.UseScratchXmm(args[0]);
const Xbyak::Xmm upper = ctx.reg_alloc.UseXmm(args[1]);
const bool part = args[2].GetImmediateU1();
code.shufpd(lower, upper, !part ? 0b00 : 0b11);
ctx.reg_alloc.DefineValue(inst, lower);
}
static void EmitVectorUnsignedAbsoluteDifference(size_t esize, EmitContext& ctx, IR::Inst* inst, BlockOfCode& code) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);

14
src/frontend/ir/ir_emitter.cpp
View file

@ -1761,6 +1761,20 @@ U128 IREmitter::VectorTableLookup(const U128& defaults, const Table& table, cons
return Inst<U128>(Opcode::VectorTableLookup128, defaults, table, indices);
}
U128 IREmitter::VectorTranspose(size_t esize, const U128& a, const U128& b, bool part) {
switch (esize) {
case 8:
return Inst<U128>(Opcode::VectorTranspose8, a, b, Imm1(part));
case 16:
return Inst<U128>(Opcode::VectorTranspose16, a, b, Imm1(part));
case 32:
return Inst<U128>(Opcode::VectorTranspose32, a, b, Imm1(part));
case 64:
return Inst<U128>(Opcode::VectorTranspose64, a, b, Imm1(part));
}
UNREACHABLE();
}
U128 IREmitter::VectorUnsignedAbsoluteDifference(size_t esize, const U128& a, const U128& b) {
switch (esize) {
case 8:

1
src/frontend/ir/ir_emitter.h
View file

@ -301,6 +301,7 @@ public:
Table VectorTable(std::vector<U128> values);
U64 VectorTableLookup(const U64& defaults, const Table& table, const U64& indices);
U128 VectorTableLookup(const U128& defaults, const Table& table, const U128& indices);
U128 VectorTranspose(size_t esize, const U128& a, const U128& b, bool part);
U128 VectorUnsignedAbsoluteDifference(size_t esize, const U128& a, const U128& b);
U128 VectorUnsignedRecipEstimate(const U128& a);
U128 VectorUnsignedRecipSqrtEstimate(const U128& a);

4
src/frontend/ir/opcodes.inc
View file

@ -473,6 +473,10 @@ OPCODE(VectorSub64, U128, U128
OPCODE(VectorTable, Table, Opaque, Opaque, Opaque, Opaque )
OPCODE(VectorTableLookup64, U64, U64, Table, U64 )
OPCODE(VectorTableLookup128, U128, U128, Table, U128 )
OPCODE(VectorTranspose8, U128, U128, U128, U1 )
OPCODE(VectorTranspose16, U128, U128, U128, U1 )
OPCODE(VectorTranspose32, U128, U128, U128, U1 )
OPCODE(VectorTranspose64, U128, U128, U128, U1 )
OPCODE(VectorUnsignedAbsoluteDifference8, U128, U128, U128 )
OPCODE(VectorUnsignedAbsoluteDifference16, U128, U128, U128 )
OPCODE(VectorUnsignedAbsoluteDifference32, U128, U128, U128 )