[llvm] Revert "[libc][bazel] Enable software prefetching for memcpy" (PR #111370)

[via llvm-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-libc

Author: Guillaume Chatelet (gchatelet)

<details>
<summary>Changes</summary>

Reverts llvm/llvm-project#<!-- -->108939

When `AVX` is available but `-mprefer-vector-width=128` some of the `mov` instructions turn into the x86 `rep;movsb` instruction leading to poor performance on "old" architectures (sandybridge, haswell). The possible solutions are : get rid of the `-mprefer-vector-width` option or use smaller static copy sizes in `inline_memcpy_x86_sse2_ge64_sw_prefetching`. Right now a copy size of 3 cache lines (192B) relying exclusively on xmm registers gets turned into `rep;movsb`.

---
Full diff: https://github.com/llvm/llvm-project/pull/111370.diff


1 Files Affected:

- (modified) utils/bazel/llvm-project-overlay/libc/libc_configure_options.bzl (+1-1) 


``````````diff
diff --git a/utils/bazel/llvm-project-overlay/libc/libc_configure_options.bzl b/utils/bazel/llvm-project-overlay/libc/libc_configure_options.bzl
index 96d7fa86e9ddf2..f65da9e98226b6 100644
--- a/utils/bazel/llvm-project-overlay/libc/libc_configure_options.bzl
+++ b/utils/bazel/llvm-project-overlay/libc/libc_configure_options.bzl
@@ -24,7 +24,7 @@ LIBC_CONFIGURE_OPTIONS = [
     # Documentation in libc/src/string/memory_utils/...
     # "LIBC_COPT_MEMCPY_USE_EMBEDDED_TINY",
     # "LIBC_COPT_MEMCPY_X86_USE_REPMOVSB_FROM_SIZE",
-    "LIBC_COPT_MEMCPY_X86_USE_SOFTWARE_PREFETCHING",
+    # "LIBC_COPT_MEMCPY_X86_USE_SOFTWARE_PREFETCHING",
     "LIBC_COPT_MEMSET_X86_USE_SOFTWARE_PREFETCHING",
 
     # Documentation in libc/docs/dev/printf_behavior.rst

``````````

</details>


https://github.com/llvm/llvm-project/pull/111370