Add tests for sin and cos on 32-bit NaN, Inf, and denorm values (#702)

As a follow up to #699 this PR re-adds the NaN, Inf, and denorm tests
for sin and cos as separate test files.

This PR also adds a new feature flag for Apple M1 (`AppleM1`) to XFAIL
non Apple M1 devices, a bug being tracked by #701

---------

Co-authored-by: Tex Riddell <texr@microsoft.com>

Author: Icohedron

test/Feature/HLSLLib/cos.nan-inf-denorm.32.test

@@ -0,0 +1,70 @@
+#--- source.hlsl
+
+StructuredBuffer<float4> In : register(t0);
+
+RWStructuredBuffer<float4> Out : register(u1);
+
+[numthreads(1,1,1)]
+void main() {
+  float4 Tmp = {cos(In[0].x), cos(In[0].yzw)};
+  Out[0] = Tmp;
+  Out[1] = cos(In[1].xyzw);
+}
+
+
+//--- pipeline.yaml
+
+---
+Shaders:
+  - Stage: Compute
+    Entry: main
+    DispatchSize: [1, 1, 1]
+Buffers:
+  - Name: In
+    Format: Float32
+    Stride: 16
+    Data: [ nan, -inf, -0x1.e7d42cp-127, 0x1.e7d42cp-127, inf, nan, nan, nan ]
+    #  NaN, -Inf, -denorm, denorm, Inf, NaN, NaN, NaN
+  - Name: Out
+    Format: Float32
+    Stride: 16
+    FillSize: 32
+  - Name: ExpectedOut # The result we expect
+    Format: Float32
+    Stride: 16
+    Data: [ nan, nan, 1.0, 1.0, nan, nan, nan, nan ]
+    #  NaN, NaN, 1.0, 1.0, NaN, NaN, NaN, NaN 
+Results:
+  - Result: Test1
+    Rule: BufferFloatEpsilon
+    Epsilon: 0.0008
+    Actual: Out
+    Expected: ExpectedOut
+DescriptorSets:
+  - Resources:
+    - Name: In
+      Kind: StructuredBuffer
+      DirectXBinding:
+        Register: 0
+        Space: 0
+      VulkanBinding:
+        Binding: 0
+    - Name: Out
+      Kind: RWStructuredBuffer
+      DirectXBinding:
+        Register: 1
+        Space: 0
+      VulkanBinding:
+        Binding: 1
+...
+#--- end
+
+# Bug https://github.com/KhronosGroup/SPIRV-Cross/issues/2525
+# XFAIL: Vulkan && MoltenVK
+
+# Bug: https://github.com/llvm/offload-test-suite/issues/701
+# XFAIL: Metal && !AppleM1 && Clang
+
+# RUN: split-file %s %t
+# RUN: %dxc_target -Gis -T cs_6_5 -Fo %t.o %t/source.hlsl
+# RUN: %offloader %t/pipeline.yaml %t.o

test/Feature/HLSLLib/sin.nan-inf-denorm.32.test

@@ -0,0 +1,70 @@
+#--- source.hlsl
+
+StructuredBuffer<float4> In : register(t0);
+
+RWStructuredBuffer<float4> Out : register(u1);
+
+[numthreads(1,1,1)]
+void main() {
+  float4 Tmp = {sin(In[0].x), sin(In[0].yzw)};
+  Out[0] = Tmp;
+  Out[1] = sin(In[1].xyzw);
+}
+
+
+//--- pipeline.yaml
+
+---
+Shaders:
+  - Stage: Compute
+    Entry: main
+    DispatchSize: [1, 1, 1]
+Buffers:
+  - Name: In
+    Format: Float32
+    Stride: 16
+    Data: [ nan, -inf, -0x1.e7d42cp-127, 0x1.e7d42cp-127, inf, nan, nan, nan ]
+    #  NaN, -Inf, -denorm, denorm, Inf, NaN, Nan, NaN
+  - Name: Out
+    Format: Float32
+    Stride: 16
+    FillSize: 32
+  - Name: ExpectedOut # The result we expect
+    Format: Float32
+    Stride: 16
+    Data: [ nan, nan, -0, 0, nan, nan, nan, nan ]
+    #  NaN, NaN, -0, 0, NaN, NaN, NaN, NaN
+Results:
+  - Result: Test1
+    Rule: BufferFloatEpsilon
+    Epsilon: 0.0008
+    Actual: Out
+    Expected: ExpectedOut
+DescriptorSets:
+  - Resources:
+    - Name: In
+      Kind: StructuredBuffer
+      DirectXBinding:
+        Register: 0
+        Space: 0
+      VulkanBinding:
+        Binding: 0
+    - Name: Out
+      Kind: RWStructuredBuffer
+      DirectXBinding:
+        Register: 1
+        Space: 0
+      VulkanBinding:
+        Binding: 1
+...
+#--- end
+
+# Bug https://github.com/KhronosGroup/SPIRV-Cross/issues/2525
+# XFAIL: Vulkan && MoltenVK
+
+# Bug: https://github.com/llvm/offload-test-suite/issues/701
+# XFAIL: Metal && !AppleM1 && Clang
+
+# RUN: split-file %s %t
+# RUN: %dxc_target -Gis -T cs_6_5 -Fo %t.o %t/source.hlsl
+# RUN: %offloader %t/pipeline.yaml %t.o

test/lit.cfg.py

@@ -104,6 +104,13 @@ def setDeviceFeatures(config, device, compiler):
         config.available_features.add("AMD")
     if "Qualcomm" in device["Description"]:
         config.available_features.add("QC")
+    if "Apple M1" in device["Description"]:
+        # As tracked by issue
+        # https://github.com/llvm/offload-test-suite/issues/701, Apple M1 Macs
+        # appear to be handling NaN, Inf, and denorm 32-bit floating-point
+        # values correctly while newer SoCs are not, even with the -Gis compiler
+        # flag.
+        config.available_features.add("AppleM1")
 
     HighestShaderModel = getHighestShaderModel(device["Features"])
     if (6, 6) <= HighestShaderModel: