feat: tensile rocm 5.5 patch

base.dockerfile

@@ -5,6 +5,8 @@ FROM ubuntu:22.04 AS base
 ARG ROCM_VERSION="5.7.3"
 ARG ROCM_APT_CODE_NAME="jammy"
 ARG ROCBLAS_BRANCH="release/rocm-rel-5.7"
+ARG TENSILE_BRANCH=${ROCBLAS_BRANCH}
+ARG TENSILE_PATCH=""
 # "gfx803" or "gfx803;gfx900"
 ARG ROCBLAS_ARCH="gfx803"
 ARG DEBIAN_FRONTEND="noninteractive"
@@ -78,13 +80,22 @@ FROM base AS patched
 # Compile and install patched rocBLAS
 USER root
 WORKDIR /root
-RUN git clone -b ${ROCBLAS_BRANCH} https://github.com/ROCmSoftwarePlatform/rocBLAS.git
+RUN <<EOF
+git clone -b ${ROCBLAS_BRANCH} https://github.com/ROCmSoftwarePlatform/rocBLAS.git
+git clone -b ${TENSILE_BRANCH} https://github.com/ROCm/Tensile.git
+EOF
+
+COPY patches/ /root/patches/
+WORKDIR /root/Tensile
+# > https://github.com/ROCm/Tensile/commit/cef9de1070c8a8cfb0c9bdc81564495f9153de05
+RUN bash -c "if [[ -n \"${TENSILE_PATCH}\" ]]; then git apply /root/patches/${TENSILE_PATCH}; fi"
+
 WORKDIR /root/rocBLAS
+# > https://github.com/xuhuisheng/rocm-build/tree/master/gfx803
 RUN rm -rf library/src/blas3/Tensile/Logic/asm_full/r9nano/
-RUN sed -i -e 's/rmake.py --install/rmake.py -a="${ROCBLAS_ARCH}" --install/g' install.sh
 
 # > https://github.com/ROCm/Tensile/pull/1897
-RUN ./install.sh -a "${ROCBLAS_ARCH}" -d --cmake_install --merge-architectures --no-lazy-library-loading
+RUN ./install.sh -a "${ROCBLAS_ARCH}" -d --cmake_install --merge-architectures --no-lazy-library-loading -t /root/Tensile
 
 WORKDIR /root/rocBLAS/build/release
 COPY fix-rocblas.sh .

patches/tensile-5.5.patch

@@ -0,0 +1,25 @@
+diff --git a/Tensile/TensileCreateLibrary.py b/Tensile/TensileCreateLibrary.py
+index 308d61fd..4037ec7c 100644
+--- a/Tensile/TensileCreateLibrary.py
++++ b/Tensile/TensileCreateLibrary.py
+@@ -1078,13 +1078,18 @@ def generateLogicDataAndSolutions(logicFiles, args):
+     #   logicData[problemType] = []
+     # logicData[problemType].append((scheduleName, deviceNames, \
+     #     solutionsForSchedule, indexOrder, exactLogic, rangeLogic ))
+-
++  (archs, _) = splitArchs()
+   if globalParameters["SeparateArchitectures"] or globalParameters["LazyLibraryLoading"]:
+     if "fallback" in masterLibraries.keys():
+       for key, value in masterLibraries.items():
+         if key != "fallback":
+           value.merge(deepcopy(masterLibraries["fallback"]))
+-
++      for archName in archs:
++        archName = archName.split('-', 1)[0]
++        if archName not in masterLibraries:
++          print1("Using fallback for arch: " + archName)
++          masterLibraries[archName] = deepcopy(masterLibraries["fallback"])
++          masterLibraries[archName].version = args.version
+       masterLibraries.pop("fallback")
+ 
+     for _, masterLibrary in masterLibraries.items():