Add arguments for controlling floating point denormal mode (#7461)

* Implement -fp-denorm-mode slangc arg

* Split fp-denorm-mode into 3 args for fp16/32/64

* Remove redundant option categories

* Use emitInst for multiple of the same OpExecutionMode

* Fix formatting

* Remove -denorm any

* Re-add option categories

* emitinst for ftz

* Use enums for type text

* Remove extra categories again

* Add tests for denorm mode

* Move denorm mode to post linking

* format code (#8)

Co-authored-by: slangbot <186143334+slangbot@users.noreply.github.com>

* regenerate command line reference (#9)

Co-authored-by: slangbot <186143334+slangbot@users.noreply.github.com>

* Clean up tests

* Fix option text

* format code (#10)

Co-authored-by: slangbot <186143334+slangbot@users.noreply.github.com>

* Add tests for "any" mode

* Return "any" enum if option not set

* Simplify emission logic

* Add support for generic entrypoints

* Move denorm modes to end of CompilerOptionName enum

* format code (#11)

Co-authored-by: slangbot <186143334+slangbot@users.noreply.github.com>

* Move new enum members to before CountOf

* Add not checks to tests, fix generic test, add functionality tests

* Rename denorm to fpDenormal

* Clean up functional test

* Rename denorm test dir

* Fix formatting, regenerate cmdline ref

* Fold simple tests into functional tests, add more dxil checks

* Remove no-op DX tests, make tests more consistent

* Disable VK functionality tests that will fail on the CI configs

* Fix formatting

* Add comments to disabled tests explaining why

---------

Co-authored-by: slangbot <ellieh+slangbot@nvidia.com>
Co-authored-by: slangbot <186143334+slangbot@users.noreply.github.com>

4 files changed, 202 insertions, 0 deletions

diff --git a/tests/fp-denormal-mode/denorm-mode-fp16.slang b/tests/fp-denormal-mode/denorm-mode-fp16.slang
new file mode 100644
index 000000000..f770ccbbe
--- /dev/null
+++ b/tests/fp-denormal-mode/denorm-mode-fp16.slang
@@ -0,0 +1,54 @@
+//TEST:SIMPLE(filecheck=CHECK_ANY):-target spirv-assembly -entry computeMain -stage compute
+//TEST:SIMPLE(filecheck=CHECK_ANY):-target spirv-assembly -entry computeMain -stage compute -denorm-mode-fp16 any
+//TEST:SIMPLE(filecheck=CHECK_PRESERVE):-target spirv-assembly -entry computeMain -stage compute -denorm-mode-fp16 preserve
+//TEST:SIMPLE(filecheck=CHECK_FTZ):-target spirv-assembly -entry computeMain -stage compute -denorm-mode-fp16 ftz
+
+//TEST:SIMPLE(filecheck=CHECK_DXIL):-target dxil-assembly -entry computeMain -stage compute -profile cs_6_2
+//TEST:SIMPLE(filecheck=CHECK_DXIL):-target dxil-assembly -entry computeMain -stage compute -profile cs_6_2 -denorm-mode-fp16 any
+//TEST:SIMPLE(filecheck=CHECK_DXIL):-target dxil-assembly -entry computeMain -stage compute -profile cs_6_2 -denorm-mode-fp16 preserve
+//TEST:SIMPLE(filecheck=CHECK_DXIL):-target dxil-assembly -entry computeMain -stage compute -profile cs_6_2 -denorm-mode-fp16 ftz
+
+//TEST(compute):COMPARE_COMPUTE(filecheck-buffer=PRESERVE):-vk -compute -Xslang -denorm-mode-fp16 -Xslang preserve
+// Capability shaderDenormFlushToZeroFloat16 is VK_FALSE on the Vulkan device used for CI testing, resulting in
+// runtime error VUID-RuntimeSpirv-shaderDenormFlushToZeroFloat16-06299 during CI testing
+//DISABLE_TEST(compute):COMPARE_COMPUTE(filecheck-buffer=FTZ):-vk -compute -Xslang -denorm-mode-fp16 -Xslang ftz
+
+// CHECK_ANY-NOT: DenormPreserve
+// CHECK_ANY-NOT: DenormFlushToZero
+
+// CHECK_PRESERVE: OpExecutionMode %computeMain DenormPreserve 16
+// CHECK_PRESERVE-NOT: OpExecutionMode %computeMain DenormPreserve 32
+// CHECK_PRESERVE-NOT: OpExecutionMode %computeMain DenormPreserve 64
+// CHECK_PRESERVE-NOT: DenormFlushToZero
+
+// CHECK_FTZ: OpExecutionMode %computeMain DenormFlushToZero 16
+// CHECK_FTZ-NOT: OpExecutionMode %computeMain DenormFlushToZero 32
+// CHECK_FTZ-NOT: OpExecutionMode %computeMain DenormFlushToZero 64
+// CHECK_FTZ-NOT: DenormPreserve
+
+// CHECK_DXIL-NOT: fp32-denorm-mode
+// CHECK_DXIL-NOT: preserve
+// CHECK_DXIL-NOT: ftz
+
+// In preserve mode, denormalized numbers should be preserved
+// PRESERVE: 66
+
+// In flush-to-zero mode, denormalized numbers should be flushed to zero
+// FTZ: 0
+
+// Smallest normal fp16
+//TEST_INPUT: set inputBuffer = ubuffer(data=[0x0400], stride=2)
+RWStructuredBuffer<half> inputBuffer;
+
+//TEST_INPUT:ubuffer(data=[0], stride=2):out,name=outputBuffer
+RWStructuredBuffer<half> outputBuffer;
+
+[shader("compute")]
+[numthreads(1, 1, 1)]
+void computeMain()
+{
+    half smallestNormal = inputBuffer[0];
+    half denormal = smallestNormal / 10;
+    
+    outputBuffer[0] = denormal;
+}
diff --git a/tests/fp-denormal-mode/denorm-mode-fp32.slang b/tests/fp-denormal-mode/denorm-mode-fp32.slang
new file mode 100644
index 000000000..530d99b6d
--- /dev/null
+++ b/tests/fp-denormal-mode/denorm-mode-fp32.slang
@@ -0,0 +1,64 @@
+//TEST:SIMPLE(filecheck=CHECK_ANY):-target spirv-assembly -entry computeMain -stage compute
+//TEST:SIMPLE(filecheck=CHECK_ANY):-target spirv-assembly -entry computeMain -stage compute -denorm-mode-fp16 any
+//TEST:SIMPLE(filecheck=CHECK_PRESERVE_SPIRV):-target spirv-assembly -entry computeMain -stage compute -denorm-mode-fp32 preserve
+//TEST:SIMPLE(filecheck=CHECK_FTZ_SPIRV):-target spirv-assembly -entry computeMain -stage compute -denorm-mode-fp32 ftz
+
+//TEST:SIMPLE(filecheck=CHECK_ANY):-target dxil-assembly -entry computeMain -stage compute -profile cs_6_2
+//TEST:SIMPLE(filecheck=CHECK_ANY):-target dxil-assembly -entry computeMain -stage compute -profile cs_6_2 -denorm-mode-fp32 any
+//TEST:SIMPLE(filecheck=CHECK_PRESERVE_DXIL):-target dxil-assembly -entry computeMain -stage compute -profile cs_6_2 -denorm-mode-fp32 preserve
+//TEST:SIMPLE(filecheck=CHECK_FTZ_DXIL):-target dxil-assembly -entry computeMain -stage compute -profile cs_6_2 -denorm-mode-fp32 ftz
+
+// Capability shaderDenormPreserveFloat32 is VK_FALSE on the Vulkan device used for CI testing, resulting in
+// runtime error VUID-RuntimeSpirv-shaderDenormPreserveFloat32-06297 during CI testing
+//DISABLE_TEST(compute):COMPARE_COMPUTE(filecheck-buffer=PRESERVE):-vk -compute -Xslang -denorm-mode-fp32 -Xslang preserve
+// Capability shaderDenormFlushToZeroFloat32 is VK_FALSE on the Vulkan device used for CI testing, resulting in
+// runtime error VUID-RuntimeSpirv-shaderDenormFlushToZeroFloat32-06300 during CI testing
+//DISABLE_TEST(compute):COMPARE_COMPUTE(filecheck-buffer=FTZ):-vk -compute -Xslang -denorm-mode-fp32 -Xslang ftz
+
+//TEST(compute):COMPARE_COMPUTE(filecheck-buffer=PRESERVE):-slang -compute -dx12 -use-dxil -profile cs_6_2 -shaderobj -Xslang -denorm-mode-fp32 -Xslang preserve
+//TEST(compute):COMPARE_COMPUTE(filecheck-buffer=FTZ):-slang -compute -dx12 -use-dxil -profile cs_6_2 -Xslang -denorm-mode-fp32 -Xslang ftz
+
+// CHECK_ANY-NOT: DenormPreserve
+// CHECK_ANY-NOT: DenormFlushToZero
+
+// CHECK_PRESERVE_SPIRV: OpExecutionMode %computeMain DenormPreserve 32
+// CHECK_PRESERVE_SPIRV-NOT: OpExecutionMode %computeMain DenormPreserve 16
+// CHECK_PRESERVE_SPIRV-NOT: OpExecutionMode %computeMain DenormPreserve 64
+// CHECK_PRESERVE_SPIRV-NOT: DenormFlushToZero
+
+// CHECK_FTZ_SPIRV: OpExecutionMode %computeMain DenormFlushToZero 32
+// CHECK_FTZ_SPIRV-NOT: OpExecutionMode %computeMain DenormFlushToZero 16
+// CHECK_FTZ_SPIRV-NOT: OpExecutionMode %computeMain DenormFlushToZero 64
+// CHECK_FTZ_SPIRV-NOT: DenormPreserve
+
+// CHECK_ANY-NOT: preserve
+// CHECK_ANY-NOT: ftz
+
+// CHECK_PRESERVE_DXIL: attributes #0 = { "fp32-denorm-mode"="preserve" }
+// CHECK_PRESERVE_DXIL-NOT: ftz
+
+// CHECK_FTZ_DXIL: attributes #0 = { "fp32-denorm-mode"="ftz" }
+// CHECK_FTZ_DXIL-NOT: preserve
+
+// In preserve mode, denormalized numbers should be preserved
+// PRESERVE: CCCCD
+
+// In flush-to-zero mode, denormalized numbers should be flushed to zero
+// FTZ: 0
+
+// Smallest normal fp32
+//TEST_INPUT: set inputBuffer = ubuffer(data=[0x00800000], stride=4)
+RWStructuredBuffer<float> inputBuffer;
+
+//TEST_INPUT:ubuffer(data=[0], stride=4):out,name=outputBuffer
+RWStructuredBuffer<float> outputBuffer;
+
+[shader("compute")]
+[numthreads(1, 1, 1)]
+void computeMain()
+{
+    float smallestNormal = inputBuffer[0];
+    float denormal = smallestNormal / 10;
+    
+    outputBuffer[0] = denormal;
+}
diff --git a/tests/fp-denormal-mode/denorm-mode-fp64.slang b/tests/fp-denormal-mode/denorm-mode-fp64.slang
new file mode 100644
index 000000000..0f4ac6c3c
--- /dev/null
+++ b/tests/fp-denormal-mode/denorm-mode-fp64.slang
@@ -0,0 +1,58 @@
+//TEST:SIMPLE(filecheck=CHECK_ANY):-target spirv-assembly -entry computeMain -stage compute
+//TEST:SIMPLE(filecheck=CHECK_ANY):-target spirv-assembly -entry computeMain -stage compute -denorm-mode-fp64 any
+//TEST:SIMPLE(filecheck=CHECK_PRESERVE):-target spirv-assembly -entry computeMain -stage compute -denorm-mode-fp64 preserve
+//TEST:SIMPLE(filecheck=CHECK_FTZ):-target spirv-assembly -entry computeMain -stage compute -denorm-mode-fp64 ftz
+
+//TEST:SIMPLE(filecheck=CHECK_DXIL):-target dxil-assembly -entry computeMain -stage compute -profile cs_6_2
+//TEST:SIMPLE(filecheck=CHECK_DXIL):-target dxil-assembly -entry computeMain -stage compute -profile cs_6_2 -denorm-mode-fp64 any
+//TEST:SIMPLE(filecheck=CHECK_DXIL):-target dxil-assembly -entry computeMain -stage compute -profile cs_6_2 -denorm-mode-fp64 preserve
+//TEST:SIMPLE(filecheck=CHECK_DXIL):-target dxil-assembly -entry computeMain -stage compute -profile cs_6_2 -denorm-mode-fp64 ftz
+
+// Capability shaderDenormPreserveFloat64 is VK_FALSE on the Vulkan device used for CI testing, resulting in
+// runtime error VUID-RuntimeSpirv-shaderDenormPreserveFloat64-06298 during CI testing
+//DISABLE_TEST(compute):COMPARE_COMPUTE(filecheck-buffer=PRESERVE):-vk -compute -Xslang -denorm-mode-fp64 -Xslang preserve
+// Capability shaderDenormFlushToZeroFloat64 is VK_FALSE on the Vulkan device used for CI testing, resulting in
+// runtime error VUID-RuntimeSpirv-shaderDenormFlushToZeroFloat64-06301 during CI testing
+//DISABLE_TEST(compute):COMPARE_COMPUTE(filecheck-buffer=FTZ):-vk -compute -Xslang -denorm-mode-fp64 -Xslang ftz
+
+// CHECK_ANY-NOT: DenormPreserve
+// CHECK_ANY-NOT: DenormFlushToZero
+
+// CHECK_PRESERVE: OpExecutionMode %computeMain DenormPreserve 64
+// CHECK_PRESERVE-NOT: OpExecutionMode %computeMain DenormPreserve 16
+// CHECK_PRESERVE-NOT: OpExecutionMode %computeMain DenormPreserve 32
+// CHECK_PRESERVE-NOT: DenormFlushToZero
+
+// CHECK_FTZ: OpExecutionMode %computeMain DenormFlushToZero 64
+// CHECK_FTZ-NOT: OpExecutionMode %computeMain DenormFlushToZero 16
+// CHECK_FTZ-NOT: OpExecutionMode %computeMain DenormFlushToZero 32
+// CHECK_FTZ-NOT: DenormPreserve
+
+// CHECK_DXIL-NOT: fp32-denorm-mode
+// CHECK_DXIL-NOT: preserve
+// CHECK_DXIL-NOT: ftz
+
+// In preserve mode, denormalized numbers should be preserved
+// PRESERVE: 9999999A
+// PRESERVE: 19999
+
+// In flush-to-zero mode, denormalized numbers should be flushed to zero
+// FTZ: 0
+// FTZ: 0
+
+// Smallest normal fp32
+//TEST_INPUT: set inputBuffer = ubuffer(data=[0x00100000 0x00000000], stride=8)
+RWStructuredBuffer<uint32_t> inputBuffer;
+
+//TEST_INPUT:ubuffer(data=[0], stride=8):out,name=outputBuffer
+RWStructuredBuffer<double> outputBuffer;
+
+[shader("compute")]
+[numthreads(1, 1, 1)]
+void computeMain()
+{
+    double smallNormal = asdouble(inputBuffer[1], inputBuffer[0]); 
+    double denorm = smallNormal / 10;
+    
+    outputBuffer[0] = denorm;
+}
diff --git a/tests/fp-denormal-mode/denorm-mode-generic.slang b/tests/fp-denormal-mode/denorm-mode-generic.slang
new file mode 100644
index 000000000..e53478d41
--- /dev/null
+++ b/tests/fp-denormal-mode/denorm-mode-generic.slang
@@ -0,0 +1,26 @@
+//TEST:SIMPLE(filecheck=CHECK_ANY):-target spirv-assembly -entry genericEntry<half> -stage compute
+//TEST:SIMPLE(filecheck=CHECK_ANY):-target spirv-assembly -entry genericEntry<float> -stage compute
+//TEST:SIMPLE(filecheck=CHECK_ANY):-target spirv-assembly -entry genericEntry<double> -stage compute
+
+//TEST:SIMPLE(filecheck=CHECK_PRESERVE):-target spirv-assembly -entry genericEntry<half> -stage compute -denorm-mode-fp32 preserve
+//TEST:SIMPLE(filecheck=CHECK_PRESERVE):-target spirv-assembly -entry genericEntry<float> -stage compute -denorm-mode-fp32 preserve
+//TEST:SIMPLE(filecheck=CHECK_PRESERVE):-target spirv-assembly -entry genericEntry<double> -stage compute -denorm-mode-fp32 preserve
+
+//TEST:SIMPLE(filecheck=CHECK_FTZ):-target spirv-assembly -entry genericEntry<half> -stage compute -denorm-mode-fp32 ftz
+//TEST:SIMPLE(filecheck=CHECK_FTZ):-target spirv-assembly -entry genericEntry<float> -stage compute -denorm-mode-fp32 ftz
+//TEST:SIMPLE(filecheck=CHECK_FTZ):-target spirv-assembly -entry genericEntry<double> -stage compute -denorm-mode-fp32 ftz
+
+// CHECK_ANY-NOT: DenormPreserve
+// CHECK_ANY-NOT: DenormFlushToZero
+
+// CHECK_PRESERVE: DenormPreserve 32
+// CHECK_FTZ: DenormFlushToZero 32
+
+[shader("compute")]  
+[numthreads(1, 1, 1)]
+__generic<T : __BuiltinArithmeticType> void genericEntry()
+{
+    T a = T(1);
+    T b = T(2);  
+    T result = a + b;
+}