Merge pull request #1691 from billhollings/VK_KHR_shader_float_controls
Add support for the VK_KHR_shader_float_controls.
diff --git a/Docs/MoltenVK_Runtime_UserGuide.md b/Docs/MoltenVK_Runtime_UserGuide.md
index ad840f5..a8433a4 100644
--- a/Docs/MoltenVK_Runtime_UserGuide.md
+++ b/Docs/MoltenVK_Runtime_UserGuide.md
@@ -289,6 +289,7 @@
- `VK_KHR_sampler_ycbcr_conversion`
- `VK_KHR_separate_depth_stencil_layouts`
- `VK_KHR_shader_draw_parameters`
+- `VK_KHR_shader_float_controls`
- `VK_KHR_shader_float16_int8`
- `VK_KHR_shader_subgroup_extended_types` *(requires Metal 2.1 on Mac or Metal 2.2 and Apple family 4 on iOS)*
- `VK_KHR_storage_buffer_storage_class`
diff --git a/Docs/Whats_New.md b/Docs/Whats_New.md
index b54e5ab..f68de67 100644
--- a/Docs/Whats_New.md
+++ b/Docs/Whats_New.md
@@ -18,6 +18,8 @@
Released TBD
+- Add support for extensions:
+ - `VK_KHR_shader_float_controls`
- Fix occassional crash from retention of `MVKSwapchain` for future drawable presentations.
- Add `MVK_USE_CEREAL` build setting to avoid use of Cereal external library (for pipeline caching).
diff --git a/MoltenVK/MoltenVK/GPUObjects/MVKDevice.mm b/MoltenVK/MoltenVK/GPUObjects/MVKDevice.mm
index 763e639..bd42b74 100644
--- a/MoltenVK/MoltenVK/GPUObjects/MVKDevice.mm
+++ b/MoltenVK/MoltenVK/GPUObjects/MVKDevice.mm
@@ -429,23 +429,23 @@
supportedProps12.conformanceVersion.minor = 0;
supportedProps12.conformanceVersion.subminor = 0;
supportedProps12.conformanceVersion.patch = 0;
- supportedProps12.denormBehaviorIndependence = VK_SHADER_FLOAT_CONTROLS_INDEPENDENCE_32_BIT_ONLY; // VK_KHR_shader_float_controls
- supportedProps12.roundingModeIndependence = VK_SHADER_FLOAT_CONTROLS_INDEPENDENCE_32_BIT_ONLY; // VK_KHR_shader_float_controls
- supportedProps12.shaderSignedZeroInfNanPreserveFloat16 = false; // VK_KHR_shader_float_controls
- supportedProps12.shaderSignedZeroInfNanPreserveFloat32 = false; // VK_KHR_shader_float_controls
- supportedProps12.shaderSignedZeroInfNanPreserveFloat64 = false; // VK_KHR_shader_float_controls
- supportedProps12.shaderDenormPreserveFloat16 = false; // VK_KHR_shader_float_controls
- supportedProps12.shaderDenormPreserveFloat32 = false; // VK_KHR_shader_float_controls
- supportedProps12.shaderDenormPreserveFloat64 = false; // VK_KHR_shader_float_controls
- supportedProps12.shaderDenormFlushToZeroFloat16 = false; // VK_KHR_shader_float_controls
- supportedProps12.shaderDenormFlushToZeroFloat32 = false; // VK_KHR_shader_float_controls
- supportedProps12.shaderDenormFlushToZeroFloat64 = false; // VK_KHR_shader_float_controls
- supportedProps12.shaderRoundingModeRTEFloat16 = false; // VK_KHR_shader_float_controls
- supportedProps12.shaderRoundingModeRTEFloat32 = false; // VK_KHR_shader_float_controls
- supportedProps12.shaderRoundingModeRTEFloat64 = false; // VK_KHR_shader_float_controls
- supportedProps12.shaderRoundingModeRTZFloat16 = false; // VK_KHR_shader_float_controls
- supportedProps12.shaderRoundingModeRTZFloat32 = false; // VK_KHR_shader_float_controls
- supportedProps12.shaderRoundingModeRTZFloat64 = false; // VK_KHR_shader_float_controls
+ supportedProps12.denormBehaviorIndependence = VK_SHADER_FLOAT_CONTROLS_INDEPENDENCE_NONE;
+ supportedProps12.roundingModeIndependence = VK_SHADER_FLOAT_CONTROLS_INDEPENDENCE_NONE;
+ supportedProps12.shaderSignedZeroInfNanPreserveFloat16 = true;
+ supportedProps12.shaderSignedZeroInfNanPreserveFloat32 = true;
+ supportedProps12.shaderSignedZeroInfNanPreserveFloat64 = false;
+ supportedProps12.shaderDenormPreserveFloat16 = false;
+ supportedProps12.shaderDenormPreserveFloat32 = false;
+ supportedProps12.shaderDenormPreserveFloat64 = false;
+ supportedProps12.shaderDenormFlushToZeroFloat16 = false;
+ supportedProps12.shaderDenormFlushToZeroFloat32 = false;
+ supportedProps12.shaderDenormFlushToZeroFloat64 = false;
+ supportedProps12.shaderRoundingModeRTEFloat16 = false;
+ supportedProps12.shaderRoundingModeRTEFloat32 = false;
+ supportedProps12.shaderRoundingModeRTEFloat64 = false;
+ supportedProps12.shaderRoundingModeRTZFloat16 = false;
+ supportedProps12.shaderRoundingModeRTZFloat32 = false;
+ supportedProps12.shaderRoundingModeRTZFloat64 = false;
supportedProps12.maxUpdateAfterBindDescriptorsInAllPools = kMVKUndefinedLargeUInt32;
supportedProps12.shaderUniformBufferArrayNonUniformIndexingNative = false;
supportedProps12.shaderSampledImageArrayNonUniformIndexingNative = _metalFeatures.arrayOfTextures && _metalFeatures.arrayOfSamplers;
@@ -612,6 +612,27 @@
texelBuffAlignProps->pNext = pNext;
break;
}
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FLOAT_CONTROLS_PROPERTIES: {
+ auto* floatControlsProperties = (VkPhysicalDeviceFloatControlsProperties*)next;
+ floatControlsProperties->denormBehaviorIndependence = supportedProps12.denormBehaviorIndependence;
+ floatControlsProperties->roundingModeIndependence = supportedProps12.roundingModeIndependence;
+ floatControlsProperties->shaderSignedZeroInfNanPreserveFloat16 = supportedProps12.shaderSignedZeroInfNanPreserveFloat16;
+ floatControlsProperties->shaderSignedZeroInfNanPreserveFloat32 = supportedProps12.shaderSignedZeroInfNanPreserveFloat32;
+ floatControlsProperties->shaderSignedZeroInfNanPreserveFloat64 = supportedProps12.shaderSignedZeroInfNanPreserveFloat64;
+ floatControlsProperties->shaderDenormPreserveFloat16 = supportedProps12.shaderDenormPreserveFloat16;
+ floatControlsProperties->shaderDenormPreserveFloat32 = supportedProps12.shaderDenormPreserveFloat32;
+ floatControlsProperties->shaderDenormPreserveFloat64 = supportedProps12.shaderDenormPreserveFloat64;
+ floatControlsProperties->shaderDenormFlushToZeroFloat16 = supportedProps12.shaderDenormFlushToZeroFloat16;
+ floatControlsProperties->shaderDenormFlushToZeroFloat32 = supportedProps12.shaderDenormFlushToZeroFloat32;
+ floatControlsProperties->shaderDenormFlushToZeroFloat64 = supportedProps12.shaderDenormFlushToZeroFloat64;
+ floatControlsProperties->shaderRoundingModeRTEFloat16 = supportedProps12.shaderRoundingModeRTEFloat16;
+ floatControlsProperties->shaderRoundingModeRTEFloat32 = supportedProps12.shaderRoundingModeRTEFloat32;
+ floatControlsProperties->shaderRoundingModeRTEFloat64 = supportedProps12.shaderRoundingModeRTEFloat64;
+ floatControlsProperties->shaderRoundingModeRTZFloat16 = supportedProps12.shaderRoundingModeRTZFloat16;
+ floatControlsProperties->shaderRoundingModeRTZFloat32 = supportedProps12.shaderRoundingModeRTZFloat32;
+ floatControlsProperties->shaderRoundingModeRTZFloat64 = supportedProps12.shaderRoundingModeRTZFloat64;
+ break;
+ }
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FRAGMENT_SHADER_BARYCENTRIC_PROPERTIES_KHR: {
auto* barycentricProperties = (VkPhysicalDeviceFragmentShaderBarycentricPropertiesKHR*)next;
barycentricProperties->triStripVertexOrderIndependentOfProvokingVertex = false;
@@ -4568,10 +4589,9 @@
}
void MVKDevice::enableExtensions(const VkDeviceCreateInfo* pCreateInfo) {
- MVKExtensionList* pWritableExtns = (MVKExtensionList*)&_enabledExtensions;
- setConfigurationResult(pWritableExtns->enable(pCreateInfo->enabledExtensionCount,
- pCreateInfo->ppEnabledExtensionNames,
- &_physicalDevice->_supportedExtensions));
+ setConfigurationResult(_enabledExtensions.enable(pCreateInfo->enabledExtensionCount,
+ pCreateInfo->ppEnabledExtensionNames,
+ &_physicalDevice->_supportedExtensions));
}
// Create the command queues
diff --git a/MoltenVK/MoltenVK/GPUObjects/MVKShaderModule.mm b/MoltenVK/MoltenVK/GPUObjects/MVKShaderModule.mm
index f748d56..40b484f 100644
--- a/MoltenVK/MoltenVK/GPUObjects/MVKShaderModule.mm
+++ b/MoltenVK/MoltenVK/GPUObjects/MVKShaderModule.mm
@@ -447,9 +447,11 @@
compile(lock, ^{
auto mtlDev = _owner->getMTLDevice();
@synchronized (mtlDev) {
+ auto mtlCompileOptions = _owner->getDevice()->getMTLCompileOptions(shaderConversionResults.entryPoint.supportsFastMath,
+ shaderConversionResults.isPositionInvariant);
+ MVKLogInfoIf(mvkConfig().debugMode, "Compiling Metal shader%s.", mtlCompileOptions.fastMathEnabled ? " with FastMath enabled" : "");
[mtlDev newLibraryWithSource: mslSourceCode
- options: _owner->getDevice()->getMTLCompileOptions(shaderConversionResults.entryPoint.supportsFastMath,
- shaderConversionResults.isPositionInvariant)
+ options: mtlCompileOptions
completionHandler: ^(id<MTLLibrary> mtlLib, NSError* error) {
bool isLate = compileComplete(mtlLib, error);
if (isLate) { destroy(); }
diff --git a/MoltenVK/MoltenVK/Layers/MVKExtensions.def b/MoltenVK/MoltenVK/Layers/MVKExtensions.def
index 604a8cf..5192947 100644
--- a/MoltenVK/MoltenVK/Layers/MVKExtensions.def
+++ b/MoltenVK/MoltenVK/Layers/MVKExtensions.def
@@ -76,6 +76,7 @@
MVK_EXTENSION(KHR_sampler_ycbcr_conversion, KHR_SAMPLER_YCBCR_CONVERSION, DEVICE, 10.11, 8.0)
MVK_EXTENSION(KHR_separate_depth_stencil_layouts, KHR_SEPARATE_DEPTH_STENCIL_LAYOUTS, DEVICE, 10.11, 8.0)
MVK_EXTENSION(KHR_shader_draw_parameters, KHR_SHADER_DRAW_PARAMETERS, DEVICE, 10.11, 8.0)
+MVK_EXTENSION(KHR_shader_float_controls, KHR_SHADER_FLOAT_CONTROLS, DEVICE, 10.11, 8.0)
MVK_EXTENSION(KHR_shader_float16_int8, KHR_SHADER_FLOAT16_INT8, DEVICE, 10.11, 8.0)
MVK_EXTENSION(KHR_shader_subgroup_extended_types, KHR_SHADER_SUBGROUP_EXTENDED_TYPES, DEVICE, 10.14, 13.0)
MVK_EXTENSION(KHR_storage_buffer_storage_class, KHR_STORAGE_BUFFER_STORAGE_CLASS, DEVICE, 10.11, 8.0)
diff --git a/Scripts/runcts b/Scripts/runcts
index acc99d1..f408a75 100755
--- a/Scripts/runcts
+++ b/Scripts/runcts
@@ -105,7 +105,7 @@
# Additional MoltenVK configuration can be set here by
# editing below, or can be set before calling this script.
export MVK_CONFIG_RESUME_LOST_DEVICE=1
-export MVK_CONFIG_FAST_MATH_ENABLED=0
+export MVK_CONFIG_FAST_MATH_ENABLED=1
export MVK_CONFIG_USE_METAL_ARGUMENT_BUFFERS=0
export MVK_CONFIG_FORCE_LOW_POWER_GPU=0
