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

emit_x64_crc32: Further improvements to codegen

Browse source
This commit is contained in:
MerryMage 2020-06-06 17:07:03 +01:00
parent bb203429c6
commit 66a356e6cb
1 changed files with 41 additions and 39 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

78
src/backend/x64/emit_x64_crc32.cpp
View file

@ -35,30 +35,52 @@ static void EmitCRC32Castagnoli(BlockOfCode& code, EmitContext& ctx, IR::Inst* i
static void EmitCRC32ISO(BlockOfCode& code, EmitContext& ctx, IR::Inst* inst, const int data_size) { static void EmitCRC32ISO(BlockOfCode& code, EmitContext& ctx, IR::Inst* inst, const int data_size) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst); auto args = ctx.reg_alloc.GetArgumentInfo(inst);
if (code.DoesCpuSupport(Xbyak::util::Cpu::tPCLMULQDQ) && data_size <= 32) { if (code.DoesCpuSupport(Xbyak::util::Cpu::tPCLMULQDQ) && data_size < 32) {
const Xbyak::Reg32 crc = ctx.reg_alloc.UseScratchGpr(args[0]).cvt32(); const Xbyak::Reg32 crc = ctx.reg_alloc.UseScratchGpr(args[0]).cvt32();
const Xbyak::Reg64 value = ctx.reg_alloc.UseGpr(args[1]); const Xbyak::Reg64 value = ctx.reg_alloc.UseScratchGpr(args[1]);
const Xbyak::Xmm xmm_crc = ctx.reg_alloc.ScratchXmm(); const Xbyak::Xmm xmm_value = ctx.reg_alloc.ScratchXmm();
const Xbyak::Xmm xmm_const = ctx.reg_alloc.ScratchXmm();
const Xbyak::Xmm xmm_tmp = ctx.reg_alloc.ScratchXmm();
code.movdqa(xmm_const, code.MConst(xword, 0x00000001'F7011641, 0x00000001'DB710641));
code.movzx(value.cvt32(), value.changeBit(data_size));
code.xor_(value.cvt32(), crc);
code.movd(xmm_tmp, value.cvt32());
code.pslldq(xmm_tmp, (64 - data_size) / 8);
if (code.DoesCpuSupport(Xbyak::util::Cpu::tAVX)) {
code.vpclmulqdq(xmm_value, xmm_tmp, xmm_const, 0x00);
code.pclmulqdq(xmm_value, xmm_const, 0x10);
code.pxor(xmm_value, xmm_tmp);
} else {
code.movdqa(xmm_value, xmm_tmp);
code.pclmulqdq(xmm_value, xmm_const, 0x00);
code.pclmulqdq(xmm_value, xmm_const, 0x10);
code.pxor(xmm_value, xmm_tmp);
}
code.pextrd(crc, xmm_value, 2);
ctx.reg_alloc.DefineValue(inst, crc);
return;
}
if (code.DoesCpuSupport(Xbyak::util::Cpu::tPCLMULQDQ) && data_size == 32) {
const Xbyak::Reg32 crc = ctx.reg_alloc.UseScratchGpr(args[0]).cvt32();
const Xbyak::Reg32 value = ctx.reg_alloc.UseGpr(args[1]).cvt32();
const Xbyak::Xmm xmm_value = ctx.reg_alloc.ScratchXmm(); const Xbyak::Xmm xmm_value = ctx.reg_alloc.ScratchXmm();
const Xbyak::Xmm xmm_const = ctx.reg_alloc.ScratchXmm(); const Xbyak::Xmm xmm_const = ctx.reg_alloc.ScratchXmm();
code.movd(xmm_value, value.cvt32());
code.movd(xmm_crc, crc);
code.movdqa(xmm_const, code.MConst(xword, 0x00000001'F7011641, 0x00000001'DB710641)); code.movdqa(xmm_const, code.MConst(xword, 0x00000001'F7011641, 0x00000001'DB710641));
code.pxor(xmm_value, xmm_crc); code.xor_(crc, value);
code.psllq(xmm_value, 64 - data_size); code.shl(crc.cvt64(), 32);
if (data_size < 32) { code.movq(xmm_value, crc.cvt64());
code.pslldq(xmm_crc, (64 - data_size) / 8);
}
code.pclmulqdq(xmm_value, xmm_const, 0x00); code.pclmulqdq(xmm_value, xmm_const, 0x00);
code.pclmulqdq(xmm_value, xmm_const, 0x10); code.pclmulqdq(xmm_value, xmm_const, 0x10);
if (data_size < 32) {
code.pxor(xmm_value, xmm_crc);
}
code.pextrd(crc, xmm_value, 2); code.pextrd(crc, xmm_value, 2);
ctx.reg_alloc.DefineValue(inst, crc); ctx.reg_alloc.DefineValue(inst, crc);
@ -67,38 +89,18 @@ static void EmitCRC32ISO(BlockOfCode& code, EmitContext& ctx, IR::Inst* inst, co
if (code.DoesCpuSupport(Xbyak::util::Cpu::tPCLMULQDQ) && data_size == 64) { if (code.DoesCpuSupport(Xbyak::util::Cpu::tPCLMULQDQ) && data_size == 64) {
const Xbyak::Reg32 crc = ctx.reg_alloc.UseScratchGpr(args[0]).cvt32(); const Xbyak::Reg32 crc = ctx.reg_alloc.UseScratchGpr(args[0]).cvt32();
const Xbyak::Reg64 value = ctx.reg_alloc.UseScratchGpr(args[1]); const Xbyak::Reg64 value = ctx.reg_alloc.UseGpr(args[1]);
const Xbyak::Xmm xmm_tmp = ctx.reg_alloc.ScratchXmm();
const Xbyak::Xmm xmm_value = ctx.reg_alloc.ScratchXmm(); const Xbyak::Xmm xmm_value = ctx.reg_alloc.ScratchXmm();
const Xbyak::Xmm xmm_const = ctx.reg_alloc.ScratchXmm(); const Xbyak::Xmm xmm_const = ctx.reg_alloc.ScratchXmm();
code.movdqa(xmm_const, code.MConst(xword, 0xFFFFFFFF00000000, 0x00000001'63CD6124)); code.movdqa(xmm_const, code.MConst(xword, 0xb4e5b025'f7011641, 0x00000001'DB710641));
code.mov(crc, crc); code.mov(crc, crc);
code.xor_(value, crc.cvt64()); code.xor_(crc.cvt64(), value);
code.movq(xmm_value, value); code.movq(xmm_value, crc.cvt64());
if (code.DoesCpuSupport(Xbyak::util::Cpu::tAVX)) {
code.vpand(xmm_tmp, xmm_value, xmm_const);
} else {
code.movdqa(xmm_tmp, xmm_value);
code.pand(xmm_tmp, xmm_const);
}
code.psllq(xmm_value, 32);
code.pclmulqdq(xmm_value, xmm_const, 0x10);
code.movdqa(xmm_const, code.MConst(xword, 0x00000001'F7011641, 0x00000001'DB710641));
if (code.DoesCpuSupport(Xbyak::util::Cpu::tAVX)) {
code.pxor(xmm_tmp, xmm_value);
code.vpclmulqdq(xmm_value, xmm_tmp, xmm_const, 0x00);
} else {
code.pxor(xmm_value, xmm_tmp);
code.movdqa(xmm_tmp, xmm_value);
code.pclmulqdq(xmm_value, xmm_const, 0x00); code.pclmulqdq(xmm_value, xmm_const, 0x00);
}
code.pclmulqdq(xmm_value, xmm_const, 0x10); code.pclmulqdq(xmm_value, xmm_const, 0x10);
code.pxor(xmm_value, xmm_tmp);
code.pextrd(crc, xmm_value, 2); code.pextrd(crc, xmm_value, 2);