Add extra guard for loop unroll pragma on AArch64
Guard clang loop unroll pragma on builds that optimize for size.
64-bit Android builds optimize for size, thus ignoring loop unroll
pragmas. The compiler warning emitted due to the ignored pragma caused
the build to fail since -Werror is also used.
Change-Id: I387bddecfa31e502751c4886fc99fbe39e761474
diff --git a/simd/arm/common/jdsample-neon.c b/simd/arm/common/jdsample-neon.c
index b8e7ae2..0371722 100644
--- a/simd/arm/common/jdsample-neon.c
+++ b/simd/arm/common/jdsample-neon.c
@@ -111,7 +111,7 @@
unsigned outptr_offset = 1;
uint8x16x2_t output_pixels;
-#if defined(__aarch64__) && defined(__clang__)
+#if defined(__aarch64__) && defined(__clang__) && !defined(__OPTIMIZE_SIZE__)
/* Unrolling by four is beneficial on AArch64 as there are 16 additional */
/* 128-bit SIMD registers to accommodate the extra data in flight. */
#pragma clang loop unroll_count(4)
