Merge branch 'main' into add_distance_tests

Author: bob80905

.github/workflows/build-and-test-callable.yaml

@@ -123,14 +123,14 @@ jobs:
       - name: Checkout OffloadTest
         uses: actions/checkout@v4
         with:
-          repository: llvm-beanz/offload-test-suite
+          repository: llvm/offload-test-suite
           ref: ${{ inputs.OffloadTest-branch }}
           path: OffloadTest
           fetch-depth: 1
       - name: Checkout Golden Images
         uses: actions/checkout@v4
         with:
-          repository: llvm-beanz/offload-golden-images
+          repository: llvm/offload-golden-images
           ref: main
           path: golden-images
           fetch-depth: 1

docs/WSL.md

@@ -10,7 +10,7 @@ These binaries, in addition to the WSL-specific headers and libraries made avail
 1. Install the DirectX-Headers package (usually named `directx-headers-dev`) from your preferred package manager.
 This is to obtain the libraries: DirectX-Guids (`libDirectX-Guids.a`) and d3dx12-format-properties (`libd3dx12-format-properties.a`)
 Alternatively, build and install the DirectX-Headers from the [original repo](https://github.com/microsoft/DirectX-Headers) or the git submodule of this repository: `third-party/DirectX-Headers`
-1. Install the rest of the [prerequisites](https://github.com/llvm-beanz/offload-test-suite/tree/main?tab=readme-ov-file#prerequisites) and follow the [instructions](https://github.com/llvm-beanz/offload-test-suite/tree/main?tab=readme-ov-file#adding-to-llvm-build) to add the experimental runtime test suite for HLSL to an LLVM build
+1. Install the rest of the [prerequisites](https://github.com/llvm/offload-test-suite/tree/main?tab=readme-ov-file#prerequisites) and follow the [instructions](https://github.com/llvm/offload-test-suite/tree/main?tab=readme-ov-file#adding-to-llvm-build) to add the experimental runtime test suite for HLSL to an LLVM build
 
 ## Known Issues
 

lib/API/VK/Device.cpp

@@ -59,6 +59,7 @@ static VkDescriptorType getDescriptorType(const ResourceKind RK) {
 static VkBufferUsageFlagBits getFlagBits(const ResourceKind RK) {
   switch (RK) {
   case ResourceKind::Buffer:
+    return VK_BUFFER_USAGE_UNIFORM_TEXEL_BUFFER_BIT;
   case ResourceKind::RWBuffer:
     return VK_BUFFER_USAGE_STORAGE_TEXEL_BUFFER_BIT;
   case ResourceKind::ByteAddressBuffer:
@@ -67,7 +68,7 @@ static VkBufferUsageFlagBits getFlagBits(const ResourceKind RK) {
   case ResourceKind::RWStructuredBuffer:
     return VK_BUFFER_USAGE_STORAGE_BUFFER_BIT;
   case ResourceKind::ConstantBuffer:
-    return VK_BUFFER_USAGE_UNIFORM_TEXEL_BUFFER_BIT;
+    return VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT;
   }
   llvm_unreachable("All cases handled");
 }
@@ -86,6 +87,73 @@ static bool isUniform(const ResourceKind RK) {
   }
   llvm_unreachable("All cases handled");
 }
+
+static std::string getMessageSeverityString(
+    VkDebugUtilsMessageSeverityFlagBitsEXT MessageSeverity) {
+  if (MessageSeverity & VK_DEBUG_UTILS_MESSAGE_SEVERITY_ERROR_BIT_EXT)
+    return "Error";
+  if (MessageSeverity & VK_DEBUG_UTILS_MESSAGE_SEVERITY_WARNING_BIT_EXT)
+    return "Warning";
+  if (MessageSeverity & VK_DEBUG_UTILS_MESSAGE_SEVERITY_INFO_BIT_EXT)
+    return "Info";
+  if (MessageSeverity & VK_DEBUG_UTILS_MESSAGE_SEVERITY_VERBOSE_BIT_EXT)
+    return "Verbose";
+  return "Unknown";
+}
+
+static VkBool32
+debugCallback(VkDebugUtilsMessageSeverityFlagBitsEXT MessageSeverity,
+              VkDebugUtilsMessageTypeFlagsEXT MessageType,
+              const VkDebugUtilsMessengerCallbackDataEXT *Data, void *) {
+  // Only interested in messages from the validation layers.
+  if (!(MessageType & VK_DEBUG_UTILS_MESSAGE_TYPE_VALIDATION_BIT_EXT))
+    return VK_FALSE;
+
+  llvm::dbgs() << "Validation " << getMessageSeverityString(MessageSeverity);
+  llvm::dbgs() << ": [ " << Data->pMessageIdName << " ]\n";
+  llvm::dbgs() << Data->pMessage;
+
+  for (uint32_t I = 0; I < Data->objectCount; I++) {
+    llvm::dbgs() << '\n';
+    if (Data->pObjects[I].pObjectName) {
+      llvm::dbgs() << "[" << Data->pObjects[I].pObjectName << "]";
+    }
+  }
+  llvm::dbgs() << '\n';
+
+  // Return true to turn the validation error or warning into an error in the
+  // vulkan API. This should causes tests to fail.
+  const bool IsErrorOrWarning =
+      MessageSeverity & (VK_DEBUG_UTILS_MESSAGE_SEVERITY_ERROR_BIT_EXT |
+                         VK_DEBUG_UTILS_MESSAGE_SEVERITY_WARNING_BIT_EXT);
+  if (IsErrorOrWarning)
+    return VK_TRUE;
+
+  // Continue to run even with VERBOSE and INFO messages.
+  return VK_FALSE;
+}
+
+static VkDebugUtilsMessengerEXT registerDebugUtilCallback(VkInstance Instance) {
+  VkDebugUtilsMessengerCreateInfoEXT CreateInfo = {};
+  CreateInfo.sType = VK_STRUCTURE_TYPE_DEBUG_UTILS_MESSENGER_CREATE_INFO_EXT;
+  CreateInfo.messageSeverity = VK_DEBUG_UTILS_MESSAGE_SEVERITY_VERBOSE_BIT_EXT |
+                               VK_DEBUG_UTILS_MESSAGE_SEVERITY_WARNING_BIT_EXT |
+                               VK_DEBUG_UTILS_MESSAGE_SEVERITY_ERROR_BIT_EXT;
+  CreateInfo.messageType = VK_DEBUG_UTILS_MESSAGE_TYPE_GENERAL_BIT_EXT |
+                           VK_DEBUG_UTILS_MESSAGE_TYPE_VALIDATION_BIT_EXT |
+                           VK_DEBUG_UTILS_MESSAGE_TYPE_PERFORMANCE_BIT_EXT;
+  CreateInfo.pfnUserCallback = debugCallback;
+  CreateInfo.pUserData = nullptr; // Optional
+  auto Func = (PFN_vkCreateDebugUtilsMessengerEXT)vkGetInstanceProcAddr(
+      Instance, "vkCreateDebugUtilsMessengerEXT");
+  if (Func == nullptr)
+    return VK_NULL_HANDLE;
+
+  VkDebugUtilsMessengerEXT DebugMessenger;
+  Func(Instance, &CreateInfo, nullptr, &DebugMessenger);
+  return DebugMessenger;
+}
+
 namespace {
 
 class VKDevice : public offloadtest::Device {
@@ -800,6 +868,7 @@ class VKDevice : public offloadtest::Device {
 class VKContext {
 private:
   VkInstance Instance = VK_NULL_HANDLE;
+  VkDebugUtilsMessengerEXT DebugMessenger = VK_NULL_HANDLE;
   llvm::SmallVector<std::shared_ptr<VKDevice>> Devices;
 
   VKContext() = default;
@@ -813,6 +882,13 @@ class VKContext {
   }
 
   void cleanup() {
+#ifndef NDEBUG
+    auto Func = (PFN_vkDestroyDebugUtilsMessengerEXT)vkGetInstanceProcAddr(
+        Instance, "vkDestroyDebugUtilsMessengerEXT");
+    if (Func != nullptr) {
+      Func(Instance, DebugMessenger, nullptr);
+    }
+#endif
     vkDestroyInstance(Instance, NULL);
     Instance = VK_NULL_HANDLE;
   }
@@ -860,6 +936,10 @@ class VKContext {
     const char *ValidationLayer = "VK_LAYER_KHRONOS_validation";
     CreateInfo.ppEnabledLayerNames = &ValidationLayer;
     CreateInfo.enabledLayerCount = 1;
+
+    const char *DebugUtilsExtensionName = "VK_EXT_debug_utils";
+    CreateInfo.ppEnabledExtensionNames = &DebugUtilsExtensionName;
+    CreateInfo.enabledExtensionCount = 1;
 #endif
 
     Res = vkCreateInstance(&CreateInfo, NULL, &Instance);
@@ -871,6 +951,10 @@ class VKContext {
                                      "Unknown Vulkan initialization error %d",
                                      Res);
 
+#ifndef NDEBUG
+    DebugMessenger = registerDebugUtilCallback(Instance);
+#endif
+
     DeviceCount = 0;
     if (vkEnumeratePhysicalDevices(Instance, &DeviceCount, nullptr))
       return llvm::createStringError(std::errc::no_such_device,

test/Basic/TestPipeline.test

@@ -40,6 +40,9 @@ DescriptorSets:
 ...
 #--- end
 
+# Tracking issue: https://github.com/llvm/llvm-project/issues/144580
+# XFAIL: Clang-Vulkan
+
 # RUN: split-file %s %t
 # RUN: %dxc_target -T cs_6_0 -E CSMain -Fo %t.o %t/source.hlsl
 # RUN: %offloader %t/pipeline.yaml %t.o | FileCheck %s

test/Feature/CBuffer/lit.local.cfg

@@ -2,6 +2,6 @@ if 'Clang' in config.available_features:
     config.unsupported = True
 
 # CBuffer bindings seem to be broken under metal
-# https://github.com/llvm-beanz/offload-test-suite/issues/55
+# https://github.com/llvm/offload-test-suite/issues/55
 if 'Metal' in config.available_features:
     config.unsupported = True

test/Feature/HLSLLib/acos.16.test

@@ -0,0 +1,67 @@
+#--- source.hlsl
+
+StructuredBuffer<half4> In : register(t0);
+
+RWStructuredBuffer<half4> Out : register(u1);
+
+[numthreads(1,1,1)]
+void main() {
+  Out[0] = acos(In[0]);
+  half4 Tmp = {acos(In[1].xyz), acos(In[1].w)};
+  Out[1] = Tmp;
+  half4 Tmp2 = {acos(In[2].xy), acos(In[2].zw)};
+  Out[2] = Tmp2;
+}
+
+//--- pipeline.yaml
+
+---
+Shaders:
+  - Stage: Compute
+    Entry: main
+    DispatchSize: [1, 1, 1]
+Buffers:
+  - Name: In
+    Format: Float16
+    Stride: 8
+    Data: [0x7e00, 0xfc00, 0x8001, 0x8000, 0x0, 0x0001, 0x7c00, 0x3c00, 0xbc00, 0x3e00, 0xbe00, 0x7e00] 
+    # nan, -inf, -denorm, -0, 0, denorm, inf, 1, -1, 1.5, -1.5, nan (filler)
+  - Name: Out
+    Format: Float16
+    Stride: 8
+    ZeroInitSize: 24
+  - Name: ExpectedOut # The result we expect
+    Format: Float16
+    Stride: 8
+    Data: [0x7e00, 0x7e00, 0x3e48, 0x3e48, 0x3e48, 0x3e48, 0x7e00, 0x0, 0x4248, 0x7e00, 0x7e00, 0x7e00]
+    # nan, nan, 1.570796, 1.570796, 1.570796, 1.570796, nan, 0, 3.1415926, nan, nan, nan
+Results:
+  - Result: Test1
+    Rule: BufferFloatEpsilon
+    Epsilon: 0.0010
+    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
+
+# REQUIRES: Half
+
+# RUN: split-file %s %t
+# RUN: %dxc_target -enable-16bit-types -T cs_6_5 -Fo %t.o %t/source.hlsl
+# RUN: %offloader %t/pipeline.yaml %t.o

test/Feature/HLSLLib/acos.32.test

@@ -0,0 +1,65 @@
+#--- source.hlsl
+
+StructuredBuffer<float4> In : register(t0);
+
+RWStructuredBuffer<float4> Out : register(u1);
+
+[numthreads(1,1,1)]
+void main() {
+  Out[0] = acos(In[0]);
+  float4 Tmp = {acos(In[1].xyz), acos(In[1].w)};
+  Out[1] = Tmp;
+  float4 Tmp2 = {acos(In[2].xy), acos(In[2].zw)};
+  Out[2] = Tmp2;
+}
+
+//--- pipeline.yaml
+
+---
+Shaders:
+  - Stage: Compute
+    Entry: main
+    DispatchSize: [1, 1, 1]
+Buffers:
+  - Name: In
+    Format: Float32
+    Stride: 16
+    Data: [nan, -inf, -0x1.e7d42cp-127, -0, 0, 0x1.e7d42cp-127, inf, 1, -1, 1.5, -1.5, nan]
+    # nan, -inf, -denorm, -0, 0, denorm, inf, 1, -1, 1.5, -1.5, nan
+  - Name: Out
+    Format: Float32
+    Stride: 16
+    ZeroInitSize: 48
+  - Name: ExpectedOut # The result we expect
+    Format: Float32
+    Stride: 16
+    Data: [nan, nan, 0x1.921fb6p+0, 0x1.921fb6p+0, 0x1.921fb6p+0, 0x1.921fb6p+0, nan, 0, 3.1415926, nan, nan, nan]
+    #[nan, nan, 1.570796, 1.570796, 1.570796, 1.570796, nan, 0, 3.14, 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
+
+# RUN: split-file %s %t
+# RUN: %dxc_target -T cs_6_5 -Fo %t.o %t/source.hlsl
+# RUN: %offloader %t/pipeline.yaml %t.o

test/Feature/HLSLLib/cos.16.test

@@ -0,0 +1,69 @@
+#--- source.hlsl
+
+StructuredBuffer<half4> In : register(t0);
+
+RWStructuredBuffer<half4> Out : register(u1);
+
+[numthreads(1,1,1)]
+void main() {
+  Out[0] = cos(In[0]);
+  half4 Tmp = {cos(In[1].xyz), cos(In[1].w)};
+  Out[1] = Tmp;
+  half4 Tmp2 = {cos(In[2].xy), cos(In[2].zw)};
+  Out[2] = Tmp2;
+  Out[3] = cos(In[3]);
+  Out[4] = cos(In[4]);
+}
+
+
+//--- pipeline.yaml
+
+---
+Shaders:
+  - Stage: Compute
+    Entry: main
+    DispatchSize: [1, 1, 1]
+Buffers:
+  - Name: In
+    Format: Float16
+    Stride: 8
+    Data: [ 0x7e00, 0xfc00, 0x8001, 0x8000, 0x0000, 0x0001, 0x7c00, 0x3906, 0x3d06, 0x3f8a, 0x4106, 0x4248, 0x438a, 0x4466, 0x4506, 0x45a7, 0x4648, 0x7e00, 0x7e00, 0x7e00,]
+    #  NaN, -Inf, -denorm, -0, 0, denorm, Inf, 0.6279297, 1.255859, 1.884766, 2.511719, 3.140625, 3.769531, 4.398438, 5.023438, 5.652344, 6.281250,
+  - Name: Out
+    Format: Float16
+    Stride: 8
+    ZeroInitSize: 40
+  - Name: ExpectedOut # The result we expect
+    Format: Float16
+    Stride: 8
+    Data: [ 0x7e00, 0x7e00, 0x3c00, 0x3c00, 0x3c00, 0x3c00, 0x7e00, 0x3a79, 0x34f5, 0xb4f1, 0xba77, 0xbc00, 0xba79, 0xb4f1, 0x34e6, 0x3a76, 0x3c00, 0x7e00, 0x7e00, 0x7e00,]
+    #  NaN, NaN, 1.0, 1.0, 1.0, 1.0, NaN, 0.80924553, 0.30975693, -0.30883664, -0.80810183, -0.99999952, -0.80924052, -0.30881903, 0.30605716, 0.80753154, 0.99999809,
+Results:
+  - Result: Test1
+    Rule: BufferFloatEpsilon
+    Epsilon: 0.003
+    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
+
+# REQUIRES: Half
+# RUN: split-file %s %t
+# RUN: %dxc_target -enable-16bit-types -T cs_6_5 -Fo %t.o %t/source.hlsl
+# RUN: %offloader %t/pipeline.yaml %t.o