Diamond75/Panda3DS
1
0
Fork 0
mirror of https://github.com/wheremyfoodat/Panda3DS.git synced 2025-04-06 22:25:41 +12:00
Code Issues Projects Releases Packages Wiki Activity Actions

Fix shader-interpreter non-IEEE outputs (#459)

Browse source 
* Re-enable non-IEEE shader test

* Fix shader-interpreter RCP/RSQ output

Handle the `-0.0` special-case

* Fix shader-interpreter MIN/MAX output

Takes advantage of min/max's properties regarding non-finites to return
NaN depending on its input position:
```
max(NaN, 2.f) -> NaN
max(2.f, NaN) -> 2
min(NaN, 2.f) -> NaN
min(2.f, NaN) -> 2
```

* Fix shader-interpreter FLR indexing bug

`3 - 1` should be `3 - i`
This commit is contained in:
Wunk 2024-03-11 23:34:02 -07:00 • committed by GitHub
parent fe9939689d
commit c89fe05b8a
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
2 changed files with 24 additions and 8 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

30
src/core/PICA/shader_interpreter.cpp
View file

@ -223,7 +223,7 @@ void PICAShader::flr(u32 instruction) {
u32 componentMask = operandDescriptor & 0xf; u32 componentMask = operandDescriptor & 0xf;
for (int i = 0; i < 4; i++) { for (int i = 0; i < 4; i++) {
if (componentMask & (1 << i)) { if (componentMask & (1 << i)) {
destVector[3 - i] = f24::fromFloat32(std::floor(srcVector[3 - 1].toFloat32())); destVector[3 - i] = f24::fromFloat32(std::floor(srcVector[3 - i].toFloat32()));
} }
} }
} }
@ -244,8 +244,12 @@ void PICAShader::max(u32 instruction) {
u32 componentMask = operandDescriptor & 0xf; u32 componentMask = operandDescriptor & 0xf;
for (int i = 0; i < 4; i++) { for (int i = 0; i < 4; i++) {
if (componentMask & (1 << i)) { if (componentMask & (1 << i)) {
const auto maximum = srcVec1[3 - i] > srcVec2[3 - i] ? srcVec1[3 - i] : srcVec2[3 - i]; const float inputA = srcVec1[3 - i].toFloat32();
destVector[3 - i] = maximum; const float inputB = srcVec2[3 - i].toFloat32();
// max(NaN, 2.f) -> NaN
// max(2.f, NaN) -> 2
const auto& maximum = std::isinf(inputB) ? inputB : std::max(inputB, inputA);
destVector[3 - i] = f24::fromFloat32(maximum);
} }
} }
} }
@ -266,8 +270,12 @@ void PICAShader::min(u32 instruction) {
u32 componentMask = operandDescriptor & 0xf; u32 componentMask = operandDescriptor & 0xf;
for (int i = 0; i < 4; i++) { for (int i = 0; i < 4; i++) {
if (componentMask & (1 << i)) { if (componentMask & (1 << i)) {
const auto mininum = srcVec1[3 - i] < srcVec2[3 - i] ? srcVec1[3 - i] : srcVec2[3 - i]; const float inputA = srcVec1[3 - i].toFloat32();
destVector[3 - i] = mininum; const float inputB = srcVec2[3 - i].toFloat32();
// min(NaN, 2.f) -> NaN
// min(2.f, NaN) -> 2
const auto& mininum = std::min(inputB, inputA);
destVector[3 - i] = f24::fromFloat32(mininum);
} }
} }
} }
@ -382,7 +390,11 @@ void PICAShader::rcp(u32 instruction) {
vec4f srcVec1 = getSourceSwizzled<1>(src1, operandDescriptor); vec4f srcVec1 = getSourceSwizzled<1>(src1, operandDescriptor);
vec4f& destVector = getDest(dest); vec4f& destVector = getDest(dest);
f24 res = f24::fromFloat32(1.0f) / srcVec1[0]; float input = srcVec1[0].toFloat32();
if (input == -0.0f) {
input = 0.0f;
}
const f24 res = f24::fromFloat32(1.0f / input);
u32 componentMask = operandDescriptor & 0xf; u32 componentMask = operandDescriptor & 0xf;
for (int i = 0; i < 4; i++) { for (int i = 0; i < 4; i++) {
@ -402,7 +414,11 @@ void PICAShader::rsq(u32 instruction) {
vec4f srcVec1 = getSourceSwizzled<1>(src1, operandDescriptor); vec4f srcVec1 = getSourceSwizzled<1>(src1, operandDescriptor);
vec4f& destVector = getDest(dest); vec4f& destVector = getDest(dest);
f24 res = f24::fromFloat32(1.0f / std::sqrt(srcVec1[0].toFloat32())); float input = srcVec1[0].toFloat32();
if (input == -0.0f) {
input = 0.0f;
}
const f24 res = f24::fromFloat32(1.0f / std::sqrt(input));
u32 componentMask = operandDescriptor & 0xf; u32 componentMask = operandDescriptor & 0xf;
for (int i = 0; i < 4; i++) { for (int i = 0; i < 4; i++) {

2
tests/shader.cpp
View file

@ -113,7 +113,7 @@ SHADER_TEST_CASE("RCP", "[shader][vertex]") {
{nihstro::OpCode::Id::END}, {nihstro::OpCode::Id::END},
}); });
// REQUIRE(shader->RunScalar({-0.0f}) == INFINITY); // Violates IEEE REQUIRE(shader->runScalar({-0.0f}) == INFINITY);
REQUIRE(shader->runScalar({0.0f}) == INFINITY); REQUIRE(shader->runScalar({0.0f}) == INFINITY);
REQUIRE(shader->runScalar({INFINITY}) == 0.0f); REQUIRE(shader->runScalar({INFINITY}) == 0.0f);
REQUIRE(std::isnan(shader->runScalar({NAN}))); REQUIRE(std::isnan(shader->runScalar({NAN})));