From faba2072ae1cf3661caceae2ef4f87cee485cf29 Mon Sep 17 00:00:00 2001 From: Benjamin Poulain Date: Mon, 13 Sep 2010 15:29:25 +0200 Subject: Use the stream operation for the SSE2 implementation of memfill32 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Writing the data of memfill() to a cacheline is unecessary because the data is not reused directly. We can use the stream operations to avoid the cache completely. When testing memfill32 separately, the function is twice as fast on Core2 and Atom. Reviewed-by: Samuel Rødal --- src/gui/painting/qdrawhelper_sse2.cpp | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/src/gui/painting/qdrawhelper_sse2.cpp b/src/gui/painting/qdrawhelper_sse2.cpp index ba2ba32..ac73958 100644 --- a/src/gui/painting/qdrawhelper_sse2.cpp +++ b/src/gui/painting/qdrawhelper_sse2.cpp @@ -266,10 +266,10 @@ void qt_memfill32_sse2(quint32 *dest, quint32 value, int count) int n = (count128 + 3) / 4; switch (count128 & 0x3) { - case 0: do { _mm_store_si128(dst128++, value128); - case 3: _mm_store_si128(dst128++, value128); - case 2: _mm_store_si128(dst128++, value128); - case 1: _mm_store_si128(dst128++, value128); + case 0: do { _mm_stream_si128(dst128++, value128); + case 3: _mm_stream_si128(dst128++, value128); + case 2: _mm_stream_si128(dst128++, value128); + case 1: _mm_stream_si128(dst128++, value128); } while (--n > 0); } -- cgit v0.12