shader: use compact integer RTZ half conversion

This commit is contained in:
nmzik
2026-10-17 12:23:12 +02:00
parent c2173801d4
commit 9f88043ebe
2 changed files with 93 additions and 17 deletions
@@ -67,6 +67,44 @@ uint32_t EmitExt(EmitterState& state, uint32_t type, uint32_t opcode,
return result;
}
uint32_t EmitF32ToF16RtzBits(EmitterState& state, uint32_t value) {
// Native half conversion causes regressions; use compact integer RTZ conversion.
const auto bits = EmitBitcastF32ToU32(state, value);
const auto abs = EmitAndConstant(state, bits, 0x7fffffffu);
const auto sign = EmitAndConstant(
state, Binary(state, spv::OpShiftRightLogical, TypeU32(state), bits, ConstantU32(state, 16)),
0x8000u);
const auto exp = Binary(state, spv::OpShiftRightLogical, TypeU32(state), abs,
ConstantU32(state, 23));
const auto mant = EmitAndConstant(state, abs, 0x007fffffu);
// Align the hidden bit to the half subnormal mantissa. Even unused shifts must be < 32.
const auto shift = EmitUMin32(
state, Binary(state, spv::OpISub, TypeU32(state), ConstantU32(state, 126), exp),
ConstantU32(state, 31));
const auto sub = Binary(state, spv::OpShiftRightLogical, TypeU32(state),
EmitOrU32(state, mant, ConstantU32(state, 0x00800000u)), shift);
// Drop 13 mantissa bits and rebias the exponent; finite overflow saturates to max half.
const auto truncated = Binary(state, spv::OpShiftRightLogical, TypeU32(state), abs,
ConstantU32(state, 13));
const auto normal = EmitUMin32(
state, Binary(state, spv::OpISub, TypeU32(state), truncated, ConstantU32(state, 0x1c000u)),
ConstantU32(state, 0x7bffu));
const auto finite = Select(state, TypeU32(state),
EmitCompareU32Constant(state, spv::OpULessThan, abs, 0x38800000u),
sub, normal);
const auto nan = EmitOrU32(state, EmitAndConstant(state, truncated, 0x03ffu),
ConstantU32(state, 0x7e00u));
const auto special = Select(state, TypeU32(state),
EmitCompareU32Constant(state, spv::OpIEqual, mant, 0),
ConstantU32(state, 0x7c00u), nan);
const auto magnitude = Select(
state, TypeU32(state),
EmitCompareU32Constant(state, spv::OpUGreaterThanEqual, abs, 0x7f800000u), special, finite);
return EmitOrU32(state, sign, magnitude);
}
uint32_t EmitF32ToU32(EmitterState& state, uint32_t src, bool signed_value) {
const auto trunc = EmitTruncF32Value(state, src);
const auto converted_raw = state.builder.AllocateId();
@@ -210,23 +248,10 @@ uint32_t EmitCompositeExtractU32x2(EmitterState& state, uint32_t arg0, IR::Value
}
uint32_t EmitPackFloat2x16Rtz(EmitterState& state, uint32_t arg0, uint32_t arg1) {
const auto pair = state.builder.AllocateId();
state.builder.AddFunction(spv::OpCompositeConstruct, TypeF32Vector(state, 2), pair, arg0, arg1);
const auto packed = EmitPackHalf2x16(state, pair);
const auto back = EmitExt(state, TypeF32Vector(state, 2), GLSLstd450UnpackHalf2x16, {packed});
auto result = packed;
for (uint32_t i = 0; i < 2; i++) {
const auto rounded = state.builder.AllocateId();
state.builder.AddFunction(spv::OpCompositeExtract, TypeF32(state), rounded, back, i);
const auto above =
Binary(state, spv::OpFOrdGreaterThan, TypeBool(state),
EmitExt(state, TypeF32(state), GLSLstd450FAbs, {rounded}),
EmitExt(state, TypeF32(state), GLSLstd450FAbs, {i == 0 ? arg0 : arg1}));
result = Binary(state, spv::OpISub, TypeU32(state), result,
Select(state, TypeU32(state), above,
ConstantU32(state, i == 0 ? 1u : 0x10000u), ConstantU32(state, 0)));
}
return result;
const auto low = EmitF32ToF16RtzBits(state, arg0);
const auto high = Binary(state, spv::OpShiftLeftLogical, TypeU32(state),
EmitF32ToF16RtzBits(state, arg1), ConstantU32(state, 16));
return EmitOrU32(state, low, high);
}
uint32_t EmitFPSaturate32(EmitterState& state, uint32_t arg0) {