Merge branch 'amd-staging' into reinstate-upstream-unrolling-patch

Author: ronlieb

.ci/generate_test_report_lib.py

@@ -92,7 +92,9 @@ def plural(num_tests):
             ]
         )
     elif failures:
-        report.extend(["", "## Failed Tests", "(click on a test name to see its output)"])
+        report.extend(
+            ["", "## Failed Tests", "(click on a test name to see its output)"]
+        )
 
         for testsuite_name, failures in failures.items():
             report.extend(["", f"### {testsuite_name}"])

.ci/metrics/metrics.py

@@ -67,6 +67,7 @@
 # remain small.
 BUILDKITE_GRAPHQL_BUILDS_PER_PAGE = 50
 
+
 @dataclass
 class JobMetrics:
     job_name: str
@@ -77,6 +78,7 @@ class JobMetrics:
     workflow_id: int
     workflow_name: str
 
+
 @dataclass
 class GaugeMetric:
     name: str
@@ -258,6 +260,7 @@ def buildkite_get_metrics(
 
     return output, incomplete_now
 
+
 def github_get_metrics(
     github_repo: github.Repository, last_workflows_seen_as_completed: set[int]
 ) -> tuple[list[JobMetrics], int]:

.github/new-prs-labeler.yml

@@ -702,6 +702,7 @@ mlgo:
   - llvm/unittests/CodeGen/ML*
   - llvm/test/CodeGen/MLRegAlloc/**
   - llvm/utils/mlgo-utils/**
+  - llvm/docs/MLGO.rst
 
 tools:llvm-exegesis:
   - llvm/tools/llvm-exegesis/**
@@ -791,6 +792,12 @@ lld:wasm:
   - lld/**/wasm/**
   - lld/Common/**
 
+backend:ARC:
+  - llvm/lib/Target/ARC/**
+  - clang/lib/Basic/Targets/ARC.h
+  - clang/lib/Basic/Targets/ARC.cpp
+  - clang/lib/CodeGen/Targets/ARC.cpp
+
 backend:ARM:
   - llvm/include/llvm/IR/IntrinsicsARM.td
   - llvm/test/MC/ARM/**
@@ -817,6 +824,19 @@ backend:AArch64:
   - clang/include/clang/Sema/SemaARM.h
   - clang/lib/Sema/SemaARM.cpp
 
+backend:CSKY:
+  - llvm/lib/Target/CSKY/**
+  - llvm/include/llvm/TargetParser/CSKYTargetParser.def
+  - llvm/include/llvm/TargetParser/CSKYTargetParser.h
+  - llvm/include/llvm/BinaryFormat/ELFRelocs/CSKY.def
+  - llvm/lib/TargetParser/CSKYTargetParser.cpp
+  - llvm/lib/Support/CSKYAttributes.cpp
+  - llvm/lib/Support/CSKYAttributeParser.cpp
+  - clang/lib/Basic/Targets/CSKY.h
+  - clang/lib/Basic/Targets/CSKY.cpp
+  - clang/lib/CodeGen/Targets/CSKY.cpp
+  - clang/lib/Driver/ToolChains/CSKY*
+
 backend:Hexagon:
   - clang/include/clang/Basic/BuiltinsHexagon*.def
   - clang/include/clang/Sema/SemaHexagon.h
@@ -840,6 +860,13 @@ backend:Hexagon:
   - llvm/test/MC/Hexagon/**
   - llvm/test/tools/llvm-objdump/ELF/Hexagon/**
 
+backend:Lanai:
+  - llvm/lib/Target/Lanai/**
+  - clang/lib/Basic/Targets/Lanai.h
+  - clang/lib/Basic/Targets/Lanai.cpp
+  - clang/lib/CodeGen/Targets/Lanai.cpp
+  - clang/lib/Driver/ToolChains/Lanai*
+
 backend:loongarch:
   - llvm/include/llvm/IR/IntrinsicsLoongArch.td
   - llvm/test/MC/LoongArch/**

.github/workflows/compute-rocm-dkmd-afar-trigger.yml

@@ -8,8 +8,9 @@ on:
 
 jobs:
   trigger_jenkins:
-    runs-on: self-hosted
-    
+    runs-on: 
+      group: compiler-generic-runners
+      
     steps:    
     - name: Set environment variable for container image
       run: |

.github/workflows/containers/github-action-ci-windows/Dockerfile

@@ -108,7 +108,7 @@ RUN choco install -y handle
 
 RUN pip3 install pywin32 buildbot-worker==2.8.4
 
-ARG RUNNER_VERSION=2.323.0
+ARG RUNNER_VERSION=2.324.0
 ENV RUNNER_VERSION=$RUNNER_VERSION
 
 RUN powershell -Command \

.github/workflows/containers/github-action-ci/Dockerfile

@@ -2,7 +2,7 @@ FROM docker.io/library/ubuntu:24.04 as base
 ENV LLVM_SYSROOT=/opt/llvm
 
 FROM base as stage1-toolchain
-ENV LLVM_VERSION=20.1.1
+ENV LLVM_VERSION=20.1.4
 
 RUN apt-get update && \
     apt-get install -y \
@@ -86,7 +86,7 @@ WORKDIR /home/gha
 
 FROM ci-container as ci-container-agent
 
-ENV GITHUB_RUNNER_VERSION=2.323.0
+ENV GITHUB_RUNNER_VERSION=2.324.0
 
 RUN mkdir actions-runner && \
     cd actions-runner && \

amd/comgr/CMakeLists.txt

@@ -265,6 +265,27 @@ if (CMAKE_SIZEOF_VOID_P EQUAL 4)
   set_target_properties(amd_comgr PROPERTIES OUTPUT_NAME "amd_comgr32")
 endif()
 
+option(LLVM_BUILD_INSTRUMENTED_COVERAGE "Build LLVM and tools with Code Coverage instrumentation" Off)
+mark_as_advanced(LLVM_BUILD_INSTRUMENTED_COVERAGE)
+if(LLVM_BUILD_INSTRUMENTED_COVERAGE)
+  if(NOT LLVM_PROFILE_MERGE_POOL_SIZE)
+    # A pool size of 1-2 is probably sufficient on an SSD. 3-4 should be fine
+    # for spinning disks. Anything higher may only help on slower mediums.
+    set(LLVM_PROFILE_MERGE_POOL_SIZE "4")
+  endif()
+  if(NOT LLVM_PROFILE_FILE_PATTERN)
+    if(NOT LLVM_PROFILE_DATA_DIR)
+      file(TO_NATIVE_PATH "${LLVM_BINARY_DIR}/profiles" LLVM_PROFILE_DATA_DIR)
+    endif()
+    file(TO_NATIVE_PATH "${LLVM_PROFILE_DATA_DIR}/%${LLVM_PROFILE_MERGE_POOL_SIZE}m.profraw" LLVM_PROFILE_FILE_PATTERN)
+  endif()
+  set(INSTRUMENTED_COVERAGE_FLAGS -O0 -fprofile-instr-generate=${LLVM_PROFILE_FILE_PATTERN} -fcoverage-mapping)
+  list(APPEND AMD_COMGR_PRIVATE_COMPILE_OPTIONS ${INSTRUMENTED_COVERAGE_FLAGS})
+  list(APPEND AMD_COMGR_PUBLIC_COMPILE_OPTIONS ${INSTRUMENTED_COVERAGE_FLAGS})
+  list(APPEND AMD_COMGR_PRIVATE_LINKER_OPTIONS ${INSTRUMENTED_COVERAGE_FLAGS} -L${LLVM_LIBRARY_DIRS})
+  list(APPEND AMD_COMGR_PUBLIC_LINKER_OPTIONS ${INSTRUMENTED_COVERAGE_FLAGS} -L${LLVM_LIBRARY_DIRS})
+endif()
+
 target_compile_options(amd_comgr
   PRIVATE "${AMD_COMGR_PRIVATE_COMPILE_OPTIONS}")
 target_compile_definitions(amd_comgr

amd/comgr/README.md

@@ -20,68 +20,67 @@ branch. LLVM should be built with at least
 
 An example `bash` session to build Comgr on Linux using GNUMakefiles is:
 
-    $ LLVM_PROJECT=~/llvm-project
-    $ DEVICE_LIBS=~/llvm-project/amd/device-libs
-    $ COMGR=~/llvm-project/amd/comgr
-    $ mkdir -p "$LLVM_PROJECT/build"
-    $ cd "$LLVM_PROJECT/build"
+    $ LLVM_PROJECT=~/llvm-project/build
+    $ DEVICE_LIBS=~/llvm-project/amd/device-libs/build
+    $ mkdir -p "$LLVM_PROJECT"
+    $ cd "$LLVM_PROJECT"
     $ cmake \
         -DCMAKE_BUILD_TYPE=Release \
         -DLLVM_ENABLE_PROJECTS="llvm;clang;lld" \
         -DLLVM_TARGETS_TO_BUILD="AMDGPU;X86" \
         ../llvm
     $ make
-    $ mkdir -p "$DEVICE_LIBS/build"
-    $ cd "$DEVICE_LIBS/build"
+    $ mkdir -p "$DEVICE_LIBS"
+    $ cd "$DEVICE_LIBS"
     $ cmake \
         -DCMAKE_BUILD_TYPE=Release \
-        -DCMAKE_PREFIX_PATH="$LLVM_PROJECT/build" \
+        -DCMAKE_PREFIX_PATH="$LLVM_PROJECT" \
         ..
     $ make
-    $ mkdir -p "$COMGR/build"
-    $ cd "$COMGR/build"
+    $ cd ~/llvm-project/amd/comgr
+    $ mkdir -p build; cd build;
     $ cmake \
         -DCMAKE_BUILD_TYPE=Release \
-        -DCMAKE_PREFIX_PATH="$LLVM_PROJECT/build;$DEVICE_LIBS/build" \
+        -DCMAKE_PREFIX_PATH="$LLVM_PROJECT;$DEVICE_LIBS" \
         ..
     $ make
     $ make test
 
 The equivalent on Windows in `cmd.exe` using Visual Studio project files is:
 
-    > set LLVM_PROJECT="%HOMEPATH%\llvm-project"
-    > set DEVICE_LIBS="%HOMEPATH%\llvm-project\amd\device-libs"
-    > set COMGR="%HOMEPATH%\llvm-project\amd\comgr"
-    > mkdir "%LLVM_PROJECT%\build"
-    > cd "%LLVM_PROJECT%\build"
+    > set LLVM_PROJECT="%HOMEPATH%\llvm-project\build"
+    > set DEVICE_LIBS="%HOMEPATH%\llvm-project\amd\device-libs\build"
+    > mkdir "%LLVM_PROJECT%"
+    > cd "%LLVM_PROJECT%"
     > cmake ^
         -DLLVM_ENABLE_PROJECTS="llvm;clang;lld" ^
         -DLLVM_TARGETS_TO_BUILD="AMDGPU;X86" ^
         ..\llvm
     > msbuild /p:Configuration=Release ALL_BUILD.vcxproj
-    > mkdir "%DEVICE_LIBS%\build"
-    > cd "%DEVICE_LIBS%\build"
+    > mkdir "%DEVICE_LIBS%"
+    > cd "%DEVICE_LIBS%"
     > cmake ^
-        -DCMAKE_PREFIX_PATH="%LLVM_PROJECT%\build" ^
+        -DCMAKE_PREFIX_PATH="%LLVM_PROJECT%" ^
         ..
     > msbuild /p:Configuration=Release ALL_BUILD.vcxproj
-    > mkdir "%COMGR%\build"
-    > cd "%COMGR%\build"
+    > cd "%HOMEPATH%\llvm-project\amd\comgr"
+    > mkdir build
+    > cd build
     > cmake ^
-        -DCMAKE_PREFIX_PATH="%LLVM_PROJECT%\build;%DEVICE_LIBS%\build" ^
+        -DCMAKE_PREFIX_PATH="%LLVM_PROJECT%;%DEVICE_LIBS%" ^
         ..
     > msbuild /p:Configuration=Release ALL_BUILD.vcxproj
     > msbuild /p:Configuration=Release RUN_TESTS.vcxproj
 
-Optionally,
+**ASAN support:** Optionally,
 [AddressSanitizer](https://github.com/google/sanitizers/wiki/AddressSanitizer)
 may be enabled during development via `-DADDRESS_SANITIZER=On` during the Comgr
 `cmake` step.
 
-Comgr can be built as a static library by passing
+**Static Comgr:** Comgr can be built as a static library by passing
 `-DCOMGR_BUILD_SHARED_LIBS=OFF` during the Comgr `cmake` step.
 
-To enable SPIRV support, checkout
+**SPIRV Support:** To enable SPIRV support, checkout
 [SPIRV-LLVM-Translator](https://github.com/ROCm/SPIRV-LLVM-Translator) in
 `llvm/projects` or `llvm/tools` and build using the above instructions, with the
 exception that the `-DCMAKE_PREFIX_PATH` for llvm-project must be an install
@@ -92,6 +91,37 @@ Comgr SPIRV-related APIs can be disabled by passing
 `-DCOMGR_DISABLE_SPIRV=1` during the Comgr `cmake` step. This removes any
 dependency on LLVM SPIRV libraries or the llvm-spirv tool.
 
+**Code Coverage Instrumentation:** Comgr supports source-based [code coverage
+via clang](https://clang.llvm.org/docs/SourceBasedCodeCoverage.html), and
+leverages the same CMake variables as
+[LLVM](https://www.llvm.org/docs/CMake.html#llvm-related-variables)
+(LLVM_BUILD_INSTRUMENTED_COVERAGE, etc.).
+
+Example of insturmenting with covereage, generating profiles, and creating an
+HTML for investigation:
+
+    $ cmake -DCMAKE_STRIP="" -DLLVM_PROFILE_DATA_DIR=`pwd`/profiles \
+        -DLLVM_BUILD_INSTRUMENTED_COVERAGE=On \
+        -DCMAKE_CXX_COMPILER="$LLVM_PROJECT/bin/clang++" \
+        -DCMAKE_C_COMPILER="$LLVM_PROJECT/bin/clang" \
+        -DCMAKE_BUILD_TYPE=Release \
+        -DCMAKE_PREFIX_PATH="$LLVM_PROJECT;$DEVICE_LIBS" ..
+    $ make -j
+    $ make test test-lit
+    $ cd profiles
+    # Manually aggregate the data and create text report.
+    $ $LLVM_PROJECT/bin/llvm-profdata merge -sparse *.profraw -o \
+        comgr_test.profdata # merge and index data
+    $ $LLVM_PROJECT/bin/llvm-cov report ../libamd_comgr.so \
+        -instr-profile=comgr_test.profdata \
+        -ignore-filename-regex="[cl].*/include/*" # show test report without \
+        includes
+    # Or use python script to aggregate the data and create html report.
+    $ $LLVM_PROJECT/../llvm/utils/prepare-code-coverage-artifact.py \
+        --preserve-profiles $LLVM_PROJECT/bin/llvm-profdata \
+        $LLVM_PROJECT/bin/llvm-cov . html ../libamd_comgr.so \
+        # create html report
+
 Depending on the Code Object Manager
 ------------------------------------
 
@@ -233,3 +263,9 @@ A script at `utils/tidy-and-format.sh` can be run to help automate the task of
 ensuring all sources conform to the coding standards. To support the use of
 this script, any exceptions must be annotated in source comments, as described
 in the clang-tidy manual.
+
+Aligning with the purpose of being a stable interface into LLVM functionality,
+the core enum values (AMD\_COMGR\_LANGUAGE_\*, AMD\_COMGR\_DATA\_KIND\_\*,
+AMD\_COMGR\_ACTION\_\*, etc.) should remain consistent between versions, even if
+some enum values are deprecated and removed. This will avoid potential breakages
+and binary incompatibilities.

amd/comgr/docs/ReleaseNotes.md

@@ -60,9 +60,10 @@ Comgr Testing, Debugging, and Logging Updates
 ---------------------------------------------
 - Removed HIP\_PATH and ROCM\_PATH environment variables. These were used for
 now-removed Comgr actions, such as \*COMPILE\_SOURCE\_TO\_FATBIN.
-- Added a new Comgr LIT testing infrastrucutre, which can be found in
+- Added a new Comgr LIT testing infrastructure, which can be found in
 amd/comgr/test-lit. This will allow us to write more in-depth and targeted
 tests.
+- Added support for source-based code coverage. See README.md for more details.
 
 New Targets
 -----------

amd/comgr/src/comgr-isa-metadata.def

@@ -47,10 +47,10 @@ HANDLE_ISA("amdgcn-amd-amdhsa-", "gfx904",           false,  true, EF_AMDGPU_MAC
 HANDLE_ISA("amdgcn-amd-amdhsa-", "gfx906",            true,  true, EF_AMDGPU_MACH_AMDGCN_GFX906,          true,  true, 65536,  32,  4,   40, 1024,   16, 800, 102,    4, 256, 256)
 HANDLE_ISA("amdgcn-amd-amdhsa-", "gfx908",            true,  true, EF_AMDGPU_MACH_AMDGCN_GFX908,          true,  true, 65536,  32,  4,   40, 1024,   16, 800, 102,    4, 256, 256)
 HANDLE_ISA("amdgcn-amd-amdhsa-", "gfx909",           false,  true, EF_AMDGPU_MACH_AMDGCN_GFX909,          true,  true, 65536,  32,  4,   40, 1024,   16, 800, 102,    4, 256, 256)
-HANDLE_ISA("amdgcn-amd-amdhsa-", "gfx90a",            true,  true, EF_AMDGPU_MACH_AMDGCN_GFX90A,          true,  true, 65536,  32,  4,   40, 1024,   16, 800, 102,    4, 256, 256)
+HANDLE_ISA("amdgcn-amd-amdhsa-", "gfx90a",            true,  true, EF_AMDGPU_MACH_AMDGCN_GFX90A,          true,  true, 65536,  32,  4,   40, 1024,   16, 800, 102,    8, 512, 512)
 HANDLE_ISA("amdgcn-amd-amdhsa-", "gfx90c",           false,  true, EF_AMDGPU_MACH_AMDGCN_GFX90C,          true,  true, 65536,  32,  4,   40, 1024,   16, 800, 102,    4, 256, 256)
-HANDLE_ISA("amdgcn-amd-amdhsa-", "gfx942",            true,  true, EF_AMDGPU_MACH_AMDGCN_GFX942,          true, false, 65536,  32,  4,   40, 1024,   16, 800, 102,    4, 256, 256)
-HANDLE_ISA("amdgcn-amd-amdhsa-", "gfx950",            true,  true, EF_AMDGPU_MACH_AMDGCN_GFX950,          true, false, 65536,  32,  4,   40, 1024,   16, 800, 102,    4, 256, 256)
+HANDLE_ISA("amdgcn-amd-amdhsa-", "gfx942",            true,  true, EF_AMDGPU_MACH_AMDGCN_GFX942,          true, false, 65536,  32,  4,   40, 1024,   16, 800, 102,    8, 512, 512)
+HANDLE_ISA("amdgcn-amd-amdhsa-", "gfx950",            true,  true, EF_AMDGPU_MACH_AMDGCN_GFX950,          true, false, 65536,  32,  4,   40, 1024,   16, 800, 102,    8, 512, 512)
 HANDLE_ISA("amdgcn-amd-amdhsa-", "gfx1010",          false,  true, EF_AMDGPU_MACH_AMDGCN_GFX1010,         true,  true, 65536,  32,  4,   40, 1024,  106, 800, 106,    8, 1024, 256)
 HANDLE_ISA("amdgcn-amd-amdhsa-", "gfx1011",          false,  true, EF_AMDGPU_MACH_AMDGCN_GFX1011,         true,  true, 65536,  32,  4,   40, 1024,  106, 800, 106,    8, 1024, 256)
 HANDLE_ISA("amdgcn-amd-amdhsa-", "gfx1012",          false,  true, EF_AMDGPU_MACH_AMDGCN_GFX1012,         true,  true, 65536,  32,  4,   40, 1024,  106, 800, 106,    8, 1024, 256)
@@ -74,7 +74,7 @@ HANDLE_ISA("amdgcn-amd-amdhsa-", "gfx1200",          false, false, EF_AMDGPU_MAC
 HANDLE_ISA("amdgcn-amd-amdhsa-", "gfx1201",          false, false, EF_AMDGPU_MACH_AMDGCN_GFX1201,         true,  true, 65536,  32,  4,   40, 1024,  106, 800, 106,   24, 1536, 256)
 
 HANDLE_ISA("amdgcn-amd-amdhsa-", "gfx9-generic",     false,  true, EF_AMDGPU_MACH_AMDGCN_GFX9_GENERIC,    true,  true, 65536,  32,  4,   40, 1024,   16, 800, 102,    4, 256, 256)
-HANDLE_ISA("amdgcn-amd-amdhsa-", "gfx9-4-generic",   true,   true, EF_AMDGPU_MACH_AMDGCN_GFX9_4_GENERIC,  true, false, 65536,  32,  4,   40, 1024,   16, 800, 102,    4, 256, 256)
+HANDLE_ISA("amdgcn-amd-amdhsa-", "gfx9-4-generic",   true,   true, EF_AMDGPU_MACH_AMDGCN_GFX9_4_GENERIC,  true, false, 65536,  32,  4,   40, 1024,   16, 800, 102,    8, 512, 512)
 HANDLE_ISA("amdgcn-amd-amdhsa-", "gfx10-1-generic",  false,  true, EF_AMDGPU_MACH_AMDGCN_GFX10_1_GENERIC, true,  true, 65536,  32,  4,   40, 1024,  106, 800, 106,    8, 1024, 256)
 HANDLE_ISA("amdgcn-amd-amdhsa-", "gfx10-3-generic",  false, false, EF_AMDGPU_MACH_AMDGCN_GFX10_3_GENERIC, true,  true, 65536,  32,  4,   40, 1024,  106, 800, 106,    8, 1024, 256)
 HANDLE_ISA("amdgcn-amd-amdhsa-", "gfx11-generic",    false, false, EF_AMDGPU_MACH_AMDGCN_GFX11_GENERIC,   true,  true, 65536,  32,  4,   40, 1024,  106, 800, 106,   16, 1024, 256)