using 4x runner for ppc 2

Author: sandeepgupta12

.ci/manywheel/build_cpu.sh

@@ -53,22 +53,6 @@ DEPS_SONAME=(
 
 rm -rf /usr/local/cuda*
 
-if [[ "$(uname -m)" == "ppc64le" ]]; then
-    echo "Disabling ld.gold for ppc64le"
-     # Remove gold from CFLAGS, CXXFLAGS, LDFLAGS
-    export CFLAGS="$(echo "${CFLAGS}" | sed 's/-fuse-ld=gold//g')"
-    export CXXFLAGS="$(echo "${CXXFLAGS}" | sed 's/-fuse-ld=gold//g')"
-    export LDFLAGS="$(echo "${LDFLAGS}" | sed 's/-fuse-ld=gold//g')"
-
-    # Pass to CMake
-    CMAKE_ARGS+=(
-        "-DCMAKE_EXE_LINKER_FLAGS='${LDFLAGS}'"
-        "-DCMAKE_SHARED_LINKER_FLAGS='${LDFLAGS}'"
-    )
-    # Clean previous build cache
-    rm -rf build/ CMakeCache.txt CMakeFiles/
-fi
-
 SOURCE_DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )" >/dev/null && pwd )"
 if [[ -z "$BUILD_PYTHONLESS" ]]; then
     BUILD_SCRIPT=build_common.sh

aten/src/ATen/cpu/vec/vec256/vsx/vec256_complex_double_vsx.h

@@ -478,7 +478,7 @@ class Vectorized<ComplexDbl> {
     this->store(tmp1);
     b.store(tmp2);
 
-    for (const auto i : c10::irange(Vectorized<c10::complex<float>>::size())) {
+    for (const auto i : c10::irange(Vectorized<c10::complex<double>>::size())) {
       out[i] = tmp1[i] / tmp2[i];
     }
     return loadu(out);