[AArch64] Re-enable SME only for Linux and new versions of Clang

This was previously disabled in 679e851, so re-enable it but only for Linux where SME is known to work correctly. Change-Id: I2626b03f3854b27162df1b55fc6767e02ffe318d Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/5802958 Reviewed-by: Frank Barchard <[email protected]> Reviewed-by: Justin Green <[email protected]>
lemenkov · Sep 23, 2024 · a37e6bc · a37e6bc
1 parent 8315fa1
commit a37e6bc
Show file tree

Hide file tree

Showing 3 changed files with 11 additions and 6 deletions.
diff --git a/include/libyuv/cpu_support.h b/include/libyuv/cpu_support.h
@@ -72,9 +72,11 @@ extern "C" {
 #define VISUALC_HAS_AVX2 1
 #endif  // VisualStudio >= 2012
 
-// Temporary disable SME.
-#if !defined(LIBYUV_DISABLE_SME)
-#define LIBYUV_DISABLE_SME
+// Clang 19 required for SME due to needing __arm_tpidr2_save from compiler-rt,
+// only enabled on Linux for now.
+#if !defined(LIBYUV_DISABLE_SME) && defined(__aarch64__) && \
+    defined(__gnu_linux__) && defined(__clang__) && (__clang_major__ >= 19)
+#define CLANG_HAS_SME 1
 #endif
 
 #ifdef __cplusplus

diff --git a/include/libyuv/rotate_row.h b/include/libyuv/rotate_row.h
@@ -50,7 +50,8 @@ extern "C" {
 #define HAS_TRANSPOSE4X4_32_NEON
 #endif
 
-#if !defined(LIBYUV_DISABLE_SME) && defined(__aarch64__)
+#if !defined(LIBYUV_DISABLE_SME) && defined(CLANG_HAS_SME) && \
+    defined(__aarch64__)
 #define HAS_TRANSPOSEWXH_SME
 #define HAS_TRANSPOSEUVWXH_SME
 #endif

diff --git a/source/rotate_sme.cc b/source/rotate_sme.cc
@@ -18,7 +18,8 @@ namespace libyuv {
 extern "C" {
 #endif
 
-#if !defined(LIBYUV_DISABLE_SME) && defined(__aarch64__)
+#if !defined(LIBYUV_DISABLE_SME) && defined(CLANG_HAS_SME) && \
+    defined(__aarch64__)
 
 __arm_locally_streaming __arm_new("za") void TransposeWxH_SME(
     const uint8_t* src,
@@ -164,7 +165,8 @@ __arm_locally_streaming __arm_new("za") void TransposeUVWxH_SME(
   } while (height > 0);
 }
 
-#endif  // !defined(LIBYUV_DISABLE_SME) && defined(__aarch64__)
+#endif  // !defined(LIBYUV_DISABLE_SME) && defined(CLANG_HAS_SME) &&
+        // defined(__aarch64__)
 
 #ifdef __cplusplus
 }  // extern "C"