소스 검색

gcc: AArch64 - Fix strict-align cpymem/setmem

The cpymemdi/setmemdi implementation doesn't fully support strict alignment.
Block the expansion if the alignment is less than 16 with STRICT_ALIGNMENT.
Clean up the condition when to use MOPS. (PR103100)

The original patch for GCC 12 removed MOPS & SIMD conditionals for setmem and
cpymem expansions in aarch64.md file. However, this version for GCC 11 does not
backport the SIMD & MOPS-related changes and retains the conditions in aarch64.md
file to preserve correctness and compatibility with the GCC 11 backend.

All changes and outputs have been verified by the author.

Upstream-Status: Backport [https://gcc.gnu.org/cgit/gcc/commit/?id=b9d16d8361a9e3a82a2f21e759e760d235d43322]

(From OE-Core rev: a99a65632116955dc69809a14bf536b22582de72)

Signed-off-by: Sundeep KOKKONDA <sundeep.kokkonda@windriver.com>
Signed-off-by: Steve Sakoman <steve@sakoman.com>
Sundeep KOKKONDA 1 주 전
부모
커밋
c6cd61f1a2
2개의 변경된 파일46개의 추가작업 그리고 0개의 파일을 삭제
  1. 1 0
      meta/recipes-devtools/gcc/gcc-11.5.inc
  2. 45 0
      meta/recipes-devtools/gcc/gcc/0032-gcc-aarch64-fix-strict-align-cpymem-setmem.patch

+ 1 - 0
meta/recipes-devtools/gcc/gcc-11.5.inc

@@ -60,6 +60,7 @@ SRC_URI = "\
            file://0029-Fix-install-path-of-linux64.h.patch \
            file://0030-rust-recursion-limit.patch \
            file://0031-gcc-sanitizers-fix.patch \
+           file://0032-gcc-aarch64-fix-strict-align-cpymem-setmem.patch \
            file://0001-CVE-2021-42574.patch \
            file://0002-CVE-2021-42574.patch \
            file://0003-CVE-2021-42574.patch \

+ 45 - 0
meta/recipes-devtools/gcc/gcc/0032-gcc-aarch64-fix-strict-align-cpymem-setmem.patch

@@ -0,0 +1,45 @@
+gcc: AArch64 - Fix strict-align cpymem/setmem
+
+The cpymemdi/setmemdi implementation doesn't fully support strict alignment.
+Block the expansion if the alignment is less than 16 with STRICT_ALIGNMENT.
+Clean up the condition when to use MOPS.
+
+Upstream-Status: Backport [https://gcc.gnu.org/cgit/gcc/commit/?id=b9d16d8361a9e3a82a2f21e759e760d235d43322]
+
+Signed-off-by: Wilco Dijkstra <wilco.dijkstra@arm.com>
+Signed-off-by: Sundeep KOKKONDA <sundeep.kokkonda@windriver.com>
+---
+--- a/gcc/config/aarch64/aarch64.c	2025-05-08 20:40:10.969865898 -0700
++++ b/gcc/config/aarch64/aarch64.c	2025-05-13 23:11:07.006796627 -0700
+@@ -23621,14 +23621,15 @@
+   int mode_bits;
+   rtx dst = operands[0];
+   rtx src = operands[1];
++  unsigned align = UINTVAL (operands[3]);
+   rtx base;
+   machine_mode cur_mode = BLKmode;
+ 
+   /* Only expand fixed-size copies.  */
+-  if (!CONST_INT_P (operands[2]))
++  if (!CONST_INT_P (operands[2]) || (STRICT_ALIGNMENT && align < 16))
+     return false;
+ 
+-  unsigned HOST_WIDE_INT size = INTVAL (operands[2]);
++  unsigned HOST_WIDE_INT size = UINTVAL (operands[2]);
+ 
+   /* Inline up to 256 bytes when optimizing for speed.  */
+   unsigned HOST_WIDE_INT max_copy_size = 256;
+@@ -23750,11 +23751,12 @@
+   unsigned HOST_WIDE_INT len;
+   rtx dst = operands[0];
+   rtx val = operands[2], src;
++  unsigned align = UINTVAL (operands[3]);
+   rtx base;
+   machine_mode cur_mode = BLKmode, next_mode;
+ 
+   /* We can't do anything smart if the amount to copy is not constant.  */
+-  if (!CONST_INT_P (operands[1]))
++  if (!CONST_INT_P (operands[1]) || (STRICT_ALIGNMENT && align < 16))
+     return false;
+ 
+   bool speed_p = !optimize_function_for_size_p (cfun);