sneedmca/Lime3DS
1
0
Fork 0
mirror of https://github.com/Lime3DS/Lime3DS synced 2024-12-26 17:12:37 -06:00
Code Issues Projects Releases Packages Wiki Activity Actions

Shader JIT: Fix CMP NaN behavior to match hardware

Browse source
This commit is contained in:
Yuri Kunde Schlesner 2015-08-23 21:46:36 -03:00
parent 83c214f6d8
commit 455147ee95
1 changed files with 23 additions and 8 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

31
src/video_core/shader/shader_jit_x64.cpp
View file

@ -578,27 +578,42 @@ void JitCompiler::Compile_CALLU(Instruction instr) {
}
void JitCompiler::Compile_CMP(Instruction instr) {
using Op = Instruction::Common::CompareOpType::Op;
Op op_x = instr.common.compare_op.x;
Op op_y = instr.common.compare_op.y;
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2, SRC2);
static const u8 cmp[] = { CMP_EQ, CMP_NEQ, CMP_LT, CMP_LE, CMP_NLE, CMP_NLT };
// SSE doesn't have greater-than (GT) or greater-equal (GE) comparison operators. You need to
// emulate them by swapping the lhs and rhs and using LT and LE. NLT and NLE can't be used here
// because they don't match when used with NaNs.
static const u8 cmp[] = { CMP_EQ, CMP_NEQ, CMP_LT, CMP_LE, CMP_LT, CMP_LE };
if (instr.common.compare_op.x == instr.common.compare_op.y) {
bool invert_op_x = (op_x == Op::GreaterThan || op_x == Op::GreaterEqual);
Gen::X64Reg lhs_x = invert_op_x ? SRC2 : SRC1;
Gen::X64Reg rhs_x = invert_op_x ? SRC1 : SRC2;
if (op_x == op_y) {
// Compare X-component and Y-component together
CMPPS(SRC1, R(SRC2), cmp[instr.common.compare_op.x]);
CMPPS(lhs_x, R(rhs_x), cmp[op_x]);
MOVQ_xmm(R(COND0), lhs_x);
MOVQ_xmm(R(COND0), SRC1);
MOV(64, R(COND1), R(COND0));
} else {
bool invert_op_y = (op_y == Op::GreaterThan || op_y == Op::GreaterEqual);
Gen::X64Reg lhs_y = invert_op_y ? SRC2 : SRC1;
Gen::X64Reg rhs_y = invert_op_y ? SRC1 : SRC2;
// Compare X-component
MOVAPS(SCRATCH, R(SRC1));
CMPSS(SCRATCH, R(SRC2), cmp[instr.common.compare_op.x]);
MOVAPS(SCRATCH, R(lhs_x));
CMPSS(SCRATCH, R(rhs_x), cmp[op_x]);
// Compare Y-component
CMPPS(SRC1, R(SRC2), cmp[instr.common.compare_op.y]);
CMPPS(lhs_y, R(rhs_y), cmp[op_y]);
MOVQ_xmm(R(COND0), SCRATCH);
MOVQ_xmm(R(COND1), SRC1);
MOVQ_xmm(R(COND1), lhs_y);
}
SHR(32, R(COND0), Imm8(31));