Merge from 'sycl' to 'sycl-web' (3 commits)

Author: iclsrc

.github/workflows/sycl-rel-nightly-launch.yml

@@ -17,23 +17,29 @@ jobs:
     permissions:
       actions: write
     runs-on: ubuntu-latest
+    strategy:
+      # Keep the matrix even if there is only one branch.
+      matrix:
+        branch: [sycl-rel-6_2, sycl-rel-6_3]
     steps:
-      - name: Launch
+      - name: Launch for ${{ matrix.branch }}
         env:
           GH_TOKEN: ${{ github.token }}
-        # To avoid excessive scheduled runs this script also checks if there are
-        # new commits since the last run - it checks if the latest commit is
-        # older >24h. That means the previous run already tested this commit.
+          BRANCH: ${{ matrix.branch }}
+        # To avoid excessive runs this script checks if there are new commits
+        # since the last run - it checks if the latest commit is older >24h.
+        # That means the previous run has already tested this commit.
         run: |
+          BRANCH="${{ matrix.branch }}"
           if [ "$GITHUB_EVENT_NAME" = "schedule" ]; then
-            latest_commit_time=$(curl -s -H "Authorization: token $GH_TOKEN" https://api.github.com/repos/intel/llvm/commits/sycl-rel-6_2 | jq -r '.commit.committer.date')
+            latest_commit_time=$(curl -s -H "Authorization: token $GH_TOKEN" https://api.github.com/repos/intel/llvm/commits/{$BRANCH} | jq -r '.commit.committer.date')
             echo $latest_commit_time
             latest_commit_epoch=$(date -d "$latest_commit_time" +%s)
             now_epoch=$(date +%s)
             diff=$((now_epoch - latest_commit_epoch))
             if [ "$diff" -lt 86400 ]; then
-              gh workflow run sycl-rel-nightly.yml --repo "${GITHUB_REPOSITORY}" --ref sycl-rel-6_2
+              gh workflow run sycl-rel-nightly.yml --repo "${GITHUB_REPOSITORY}" --ref "${BRANCH}"
             fi
           else
-            gh workflow run sycl-rel-nightly.yml --repo "${GITHUB_REPOSITORY}" --ref sycl-rel-6_2
+            gh workflow run sycl-rel-nightly.yml --repo "${GITHUB_REPOSITORY}" --ref "${BRANCH}"
           fi

.github/workflows/sycl-rel-nightly.yml

@@ -1,5 +1,6 @@
-# The actual version of this workflow can be found on the latest sycl-rel-*
-# branch. This file is still needed to launch the workflow on another branches.
+# The actual version of this file can be found on the latest sycl-rel-* branch.
+# A new release branch copies this file from the latest release branch.
+# This file is still needed to launch the workflow on another branches.
 
 name: SYCL Release Branch Nightly
 

llvm/lib/Analysis/ConstantFolding.cpp

@@ -1704,6 +1704,12 @@ bool llvm::canConstantFoldCallTo(const CallBase *Call, const Function *F) {
   case Intrinsic::amdgcn_fma_legacy:
   case Intrinsic::amdgcn_fract:
   case Intrinsic::amdgcn_sin:
+
+  // Floating point builtin intrinsics can be folded if accuracy requirements
+  // are satisfied in addition to the rules defined for regular floating point
+  // operations.
+  case Intrinsic::fpbuiltin_sqrt:
+
   // The intrinsics below depend on rounding mode in MXCSR.
   case Intrinsic::x86_sse_cvtss2si:
   case Intrinsic::x86_sse_cvtss2si64:
@@ -2623,6 +2629,7 @@ static Constant *ConstantFoldScalarCall1(StringRef Name,
           return ConstantFP::get(Ty->getContext(), U);
         return ConstantFoldFP(atan, APF, Ty);
       case Intrinsic::sqrt:
+      case Intrinsic::fpbuiltin_sqrt:
         return ConstantFoldFP(sqrt, APF, Ty);
 
       // NVVM Intrinsics:

llvm/test/Transforms/Inline/inline_constprop.ll

@@ -374,3 +374,18 @@ bb.true:
 bb.false:
   ret float %x
 }
+
+define float @caller9() {
+; Check that we can constant-prop through fp intrinsics.
+;
+; CHECK-LABEL: @caller9(
+; CHECK-NEXT: ret float 2.000000e+00
+  %x = call float @callee9(float 16.0)
+  ret float %x
+}
+
+define float @callee9(float %x) {
+  %s = call fast float @llvm.sqrt.f32(float %x)
+  %fs = call fast float @llvm.fpbuiltin.sqrt.f32(float %s)
+  ret float %fs
+}

sycl/source/detail/compression.hpp

@@ -33,7 +33,9 @@ class ZSTDCompressor {
 
   // Get the singleton instance of the ZSTDCompressor class.
   static ZSTDCompressor &GetSingletonInstance() {
-    static ZSTDCompressor instance;
+    // Use thread_local to ensure that each thread has its own instance.
+    // This avoids issues with concurrent access to the ZSTD contexts.
+    thread_local ZSTDCompressor instance;
     return instance;
   }
 

sycl/unittests/compression/CompressionTests.cpp

@@ -6,6 +6,7 @@
 //
 //===----------------------------------------------------------------------===//
 
+#include "../thread_safety/ThreadUtils.h"
 #include <detail/compression.hpp>
 #include <sycl/sycl.hpp>
 
@@ -79,3 +80,36 @@ TEST(CompressionTest, EmptyInputTest) {
   std::string decompressedStr((char *)decompressedData.get(), decompressedSize);
   ASSERT_EQ(input, decompressedStr);
 }
+
+// Test to check for concurrent compression and decompression.
+TEST(CompressionTest, ConcurrentCompressionDecompression) {
+  std::string data = "Concurrent compression and decompression test!";
+
+  constexpr size_t ThreadCount = 20;
+
+  Barrier b(ThreadCount);
+  {
+    auto testCompressDecompress = [&](size_t threadId) {
+      b.wait();
+      size_t compressedDataSize = 0;
+      auto compressedData = ZSTDCompressor::CompressBlob(
+          data.c_str(), data.size(), compressedDataSize, 3);
+
+      ASSERT_NE(compressedData, nullptr);
+      ASSERT_GT(compressedDataSize, (size_t)0);
+
+      size_t decompressedSize = 0;
+      auto decompressedData = ZSTDCompressor::DecompressBlob(
+          compressedData.get(), compressedDataSize, decompressedSize);
+
+      ASSERT_NE(decompressedData, nullptr);
+      ASSERT_GT(decompressedSize, (size_t)0);
+
+      std::string decompressedStr((char *)decompressedData.get(),
+                                  decompressedSize);
+      ASSERT_EQ(data, decompressedStr);
+    };
+
+    ::ThreadPool MPool(ThreadCount, testCompressDecompress);
+  }
+}