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

emit_x64_vector: GNFI+SSSE3 implementation of EmitVectorReverseBits

Browse source 
Performs a full 128-bit bit-reversal using only two instructions.

First by reversing all the bits of each byte using a galois matrix
multiplication(vgf2p8affineqb, Icelake), and then by reversing the bytes
themselves(pshufb, ssse3).
This commit is contained in:
Wunkolo 2020-09-09 10:06:33 -07:00 committed by merry
parent eb00bea1ff
commit 6bb49726f4
1 changed files with 37 additions and 29 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

66
src/backend/x64/emit_x64_vector.cpp
View file

@ -2747,40 +2747,48 @@ void EmitX64::EmitVectorReverseBits(EmitContext& ctx, IR::Inst* inst) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst); auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm data = ctx.reg_alloc.UseScratchXmm(args[0]); const Xbyak::Xmm data = ctx.reg_alloc.UseScratchXmm(args[0]);
const Xbyak::Xmm high_nibble_reg = ctx.reg_alloc.ScratchXmm();
code.movdqa(high_nibble_reg, code.MConst(xword, 0xF0F0F0F0F0F0F0F0, 0xF0F0F0F0F0F0F0F0)); if (code.HasAVX512_Icelake() && code.HasSSSE3()) {
code.pand(high_nibble_reg, data); // GFNI(vgf2p8affineqb) and SSSE3(pshuf)
code.pxor(data, high_nibble_reg); // Reverse bits within bytes
code.psrld(high_nibble_reg, 4); code.vgf2p8affineqb(data, data, code.MConst(xword_b, 0x8040201008040201), 0);
// Reverse bytes within vector
if (code.HasSSSE3()) { code.pshufb(data, code.MConst(xword, 0x0001020304050607, 0x08090a0b0c0d0e0f));
// High lookup
const Xbyak::Xmm high_reversed_reg = ctx.reg_alloc.ScratchXmm();
code.movdqa(high_reversed_reg, code.MConst(xword, 0xE060A020C0408000, 0xF070B030D0509010));
code.pshufb(high_reversed_reg, data);
// Low lookup (low nibble equivalent of the above)
code.movdqa(data, code.MConst(xword, 0x0E060A020C040800, 0x0F070B030D050901));
code.pshufb(data, high_nibble_reg);
code.por(data, high_reversed_reg);
} else { } else {
code.pslld(data, 4); const Xbyak::Xmm high_nibble_reg = ctx.reg_alloc.ScratchXmm();
code.por(data, high_nibble_reg); code.movdqa(high_nibble_reg, code.MConst(xword, 0xF0F0F0F0F0F0F0F0, 0xF0F0F0F0F0F0F0F0));
code.movdqa(high_nibble_reg, code.MConst(xword, 0xCCCCCCCCCCCCCCCC, 0xCCCCCCCCCCCCCCCC));
code.pand(high_nibble_reg, data); code.pand(high_nibble_reg, data);
code.pxor(data, high_nibble_reg); code.pxor(data, high_nibble_reg);
code.psrld(high_nibble_reg, 2); code.psrld(high_nibble_reg, 4);
code.pslld(data, 2);
code.por(data, high_nibble_reg);
code.movdqa(high_nibble_reg, code.MConst(xword, 0xAAAAAAAAAAAAAAAA, 0xAAAAAAAAAAAAAAAA)); if (code.HasSSSE3()) {
code.pand(high_nibble_reg, data); // High lookup
code.pxor(data, high_nibble_reg); const Xbyak::Xmm high_reversed_reg = ctx.reg_alloc.ScratchXmm();
code.psrld(high_nibble_reg, 1); code.movdqa(high_reversed_reg, code.MConst(xword, 0xE060A020C0408000, 0xF070B030D0509010));
code.paddd(data, data); code.pshufb(high_reversed_reg, data);
code.por(data, high_nibble_reg);
// Low lookup (low nibble equivalent of the above)
code.movdqa(data, code.MConst(xword, 0x0E060A020C040800, 0x0F070B030D050901));
code.pshufb(data, high_nibble_reg);
code.por(data, high_reversed_reg);
} else {
code.pslld(data, 4);
code.por(data, high_nibble_reg);
code.movdqa(high_nibble_reg, code.MConst(xword, 0xCCCCCCCCCCCCCCCC, 0xCCCCCCCCCCCCCCCC));
code.pand(high_nibble_reg, data);
code.pxor(data, high_nibble_reg);
code.psrld(high_nibble_reg, 2);
code.pslld(data, 2);
code.por(data, high_nibble_reg);
code.movdqa(high_nibble_reg, code.MConst(xword, 0xAAAAAAAAAAAAAAAA, 0xAAAAAAAAAAAAAAAA));
code.pand(high_nibble_reg, data);
code.pxor(data, high_nibble_reg);
code.psrld(high_nibble_reg, 1);
code.paddd(data, data);
code.por(data, high_nibble_reg);
}
} }
ctx.reg_alloc.DefineValue(inst, data); ctx.reg_alloc.DefineValue(inst, data);