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

shader_jit_a64_compiler: Improve Compile_SwizzleSrc (#7136)

Browse source
This commit is contained in:
merry 2023-11-11 12:56:48 +00:00 committed by GitHub
parent fa08df21a5
commit 80213bf88f
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
1 changed files with 28 additions and 16 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

44
src/video_core/shader/shader_jit_a64_compiler.cpp
View file

@ -257,28 +257,40 @@ void JitShader::Compile_SwizzleSrc(Instruction instr, unsigned src_num, SourceRe
// Generate instructions for source register swizzling as needed // Generate instructions for source register swizzling as needed
u8 sel = swiz.GetRawSelector(src_num); u8 sel = swiz.GetRawSelector(src_num);
if (sel != NO_SRC_REG_SWIZZLE) { switch (sel) {
case NO_SRC_REG_SWIZZLE:
// NOP
break;
case 0b00'00'00'00:
DUP(dest.S4(), dest.Selem()[0]);
break;
case 0b01'01'01'01:
DUP(dest.S4(), dest.Selem()[1]);
break;
case 0b10'10'10'10:
DUP(dest.S4(), dest.Selem()[2]);
break;
case 0b11'11'11'11:
DUP(dest.S4(), dest.Selem()[3]);
break;
default: {
const int table[] = { const int table[] = {
((sel & 0b11'00'00'00) >> 6), ((sel & 0b11'00'00'00) >> 6),
((sel & 0b00'11'00'00) >> 4), ((sel & 0b00'11'00'00) >> 4),
((sel & 0b00'00'11'00) >> 2), ((sel & 0b00'00'11'00) >> 2),
((sel & 0b00'00'00'11) >> 0), ((sel & 0b00'00'00'11) >> 0),
}; };
MOV(VSCRATCH0.B16(), dest.B16());
// Generate table-vector if (table[0] != 0)
MOV(XSCRATCH0.toW(), u32(0x03'02'01'00u + (table[0] * 0x04'04'04'04u))); MOV(dest.Selem()[0], VSCRATCH0.Selem()[table[0]]);
MOV(VSCRATCH0.Selem()[0], XSCRATCH0.toW()); if (table[1] != 1)
MOV(dest.Selem()[1], VSCRATCH0.Selem()[table[1]]);
MOV(XSCRATCH0.toW(), u32(0x03'02'01'00u + (table[1] * 0x04'04'04'04u))); if (table[2] != 2)
MOV(VSCRATCH0.Selem()[1], XSCRATCH0.toW()); MOV(dest.Selem()[2], VSCRATCH0.Selem()[table[2]]);
if (table[3] != 3)
MOV(XSCRATCH0.toW(), u32(0x03'02'01'00u + (table[2] * 0x04'04'04'04u))); MOV(dest.Selem()[3], VSCRATCH0.Selem()[table[3]]);
MOV(VSCRATCH0.Selem()[2], XSCRATCH0.toW()); break;
}
MOV(XSCRATCH0.toW(), u32(0x03'02'01'00u + (table[3] * 0x04'04'04'04u)));
MOV(VSCRATCH0.Selem()[3], XSCRATCH0.toW());
TBL(dest.B16(), List{dest.B16()}, VSCRATCH0.B16());
} }
// If the source register should be negated, flip the negative bit using XOR // If the source register should be negated, flip the negative bit using XOR