Use a wider type for local variables holding the cache line size

This allows the compiler to do the widening operation once, rather than repeatedly in loops, fixing a performance regression introduced in eclipse#6515. Signed-off-by: Keith W. Campbell <[email protected]>
keithc-ca · Jun 23, 2022 · 8ca8d51 · 8ca8d51
1 parent 8b620b8
commit 8ca8d51
Show file tree

Hide file tree

Showing 2 changed files with 6 additions and 5 deletions.
diff --git a/port/unix/omrcpu.c b/port/unix/omrcpu.c
@@ -98,7 +98,7 @@ void
 omrcpu_flush_icache(struct OMRPortLibrary *portLibrary, void *memoryPointer, uintptr_t byteAmount)
 {
 #if defined(LINUXPPC) || defined(PPC) || defined(RS6000)
-	uint32_t cacheLineSize = PPG_mem_ppcCacheLineSize;
+	uintptr_t cacheLineSize = PPG_mem_ppcCacheLineSize;
 	unsigned char *addr = NULL;
 	unsigned char *limit = (unsigned char *)
 			(((uintptr_t)memoryPointer + byteAmount + (cacheLineSize - 1))

diff --git a/util/omrutil/j9memclr.cpp b/util/omrutil/j9memclr.cpp
@@ -74,13 +74,14 @@ OMRZeroMemory(void *ptr, uintptr_t length)
 	}
 #endif /* defined(LINUXPPC) */
 
+	uintptr_t localCacheLineSize = cacheLineSize;
+
 	/* one-time-only calculation of cache line size */
-	if (0 == cacheLineSize) {
-		cacheLineSize = getCacheLineSize();
+	if (0 == localCacheLineSize) {
+		localCacheLineSize = getCacheLineSize();
+		cacheLineSize = localCacheLineSize;
 	}
 
-	uint32_t localCacheLineSize = cacheLineSize;
-
 	/* Zeroing by dcbz is effective if requested length is at least twice larger then Data Cache Block size */
 	if (length < (2 * localCacheLineSize)) {
 		memset(ptr, 0, (size_t)length);