fabi/Xenia-Canary
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[x64] Add software fallback for PACK_FLOAT16_4

Browse Source 
Current implementation has an off by 1 in rounding, should
round up to even but doesn't. Need to figure out how to implement
it properly so just leaving the software version here for later
verification.
This commit is contained in:
Herman S.
2025-10-25 21:56:13 +09:00
parent 374ec3634e
commit 1182f9d73d
1 changed files with 37 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

38
src/xenia/cpu/backend/x64/x64_seq_vector.cc
View File

@@ -2744,14 +2744,50 @@ struct PACK : Sequence<PACK, I<OPCODE_PACK, V128Op, V128Op, V128Op>> {
e.vmovaps(i.dest, e.xmm0);
}
static __m128i EmulateFLOAT16_4_RoundToEven(void*, __m128 src1) {
alignas(16) uint16_t b[8] = {0};
// Extract and convert all 4 floats
float f0 = _mm_cvtss_f32(src1);
float f1 =
_mm_cvtss_f32(_mm_shuffle_ps(src1, src1, _MM_SHUFFLE(1, 1, 1, 1)));
float f2 =
_mm_cvtss_f32(_mm_shuffle_ps(src1, src1, _MM_SHUFFLE(2, 2, 2, 2)));
float f3 =
_mm_cvtss_f32(_mm_shuffle_ps(src1, src1, _MM_SHUFFLE(3, 3, 3, 3)));
// Convert to float16 with round-to-nearest-even
// Index calculation: 7 - (i ^ 2) where ^ is XOR
b[5] = float_to_xenos_half(f0, false, true); // 7 - (0 ^ 2) = 7 - 2 = 5
b[4] = float_to_xenos_half(f1, false, true); // 7 - (1 ^ 2) = 7 - 3 = 4
b[7] = float_to_xenos_half(f2, false, true); // 7 - (2 ^ 2) = 7 - 0 = 7
b[6] = float_to_xenos_half(f3, false, true); // 7 - (3 ^ 2) = 7 - 1 = 6
return _mm_load_si128(reinterpret_cast<const __m128i*>(b));
}
static void EmitFLOAT16_4(X64Emitter& e, const EmitArgType& i) {
if (!i.src1.is_constant) {
#if XE_ARCH_AMD64
// TODO(has207): this code has an off-by-1 in rounding
// compared to the fallback which is accurate, but
// keeping it as default. Should be fixed.
emit_fast_f16_pack(e, i, XMMPackFLOAT16_4);
#else
auto src1 = GetInputRegOrConstant(e, i.src1, e.xmm3);
#if XE_PLATFORM_WIN32
e.lea(e.GetNativeParam(0), e.StashXmm(0, src1));
#else
e.vmovaps(e.xmm0, src1);
#endif
e.CallNativeSafe(reinterpret_cast<void*>(EmulateFLOAT16_4_RoundToEven));
e.vmovdqa(i.dest, e.xmm0);
#endif
} else {
vec128_t result = vec128b(0);
for (unsigned idx = 0; idx < 4; ++idx) {
result.u16[(7 - (idx ^ 2))] =
float_to_xenos_half(i.src1.constant().f32[idx]);
float_to_xenos_half(i.src1.constant().f32[idx], false, true);
}
e.LoadConstantXmm(i.dest, result);