mesa: Add a streaming load memcpy implementation.

Uses SSE 4.1's MOVNTDQA instruction (streaming load) to read from
uncached memory without polluting the cache.

Reviewed-by: Chad Versace <[email protected]>

1 files changed, 85 insertions, 0 deletions

diff --git a/src/mesa/main/streaming-load-memcpy.c b/src/mesa/main/streaming-load-memcpy.c
new file mode 100644
index 00000000000..d7147afdc5c
--- /dev/null
+++ b/src/mesa/main/streaming-load-memcpy.c
@@ -0,0 +1,85 @@
+/*
+ * Copyright © 2013 Intel Corporation
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ *
+ * Authors:
+ *    Eric Anholt <[email protected]>
+ *    Matt Turner <[email protected]>
+ *
+ */
+
+#include "main/macros.h"
+#include "main/streaming-load-memcpy.h"
+#include <smmintrin.h>
+
+/* Copies memory from src to dst, using SSE 4.1's MOVNTDQA to get streaming
+ * read performance from uncached memory.
+ */
+void
+_mesa_streaming_load_memcpy(void *restrict dst, void *restrict src, size_t len)
+{
+   char *restrict d = dst;
+   char *restrict s = src;
+
+   /* If dst and src are not co-aligned, fallback to memcpy(). */
+   if (((uintptr_t)d & 15) != ((uintptr_t)s & 15)) {
+      memcpy(d, s, len);
+      return;
+   }
+
+   /* memcpy() the misaligned header. At the end of this if block, <d> and <s>
+    * are aligned to a 16-byte boundary or <len> == 0.
+    */
+   if ((uintptr_t)d & 15) {
+      uintptr_t bytes_before_alignment_boundary = 16 - ((uintptr_t)d & 15);
+      assert(bytes_before_alignment_boundary < 16);
+
+      memcpy(d, s, MIN2(bytes_before_alignment_boundary, len));
+
+      d = (char *)ALIGN((uintptr_t)d, 16);
+      s = (char *)ALIGN((uintptr_t)s, 16);
+      len -= MIN2(bytes_before_alignment_boundary, len);
+   }
+
+   while (len >= 64) {
+      __m128i *dst_cacheline = (__m128i *)d;
+      __m128i *src_cacheline = (__m128i *)s;
+
+      __m128i temp1 = _mm_stream_load_si128(src_cacheline + 0);
+      __m128i temp2 = _mm_stream_load_si128(src_cacheline + 1);
+      __m128i temp3 = _mm_stream_load_si128(src_cacheline + 2);
+      __m128i temp4 = _mm_stream_load_si128(src_cacheline + 3);
+
+      _mm_store_si128(dst_cacheline + 0, temp1);
+      _mm_store_si128(dst_cacheline + 1, temp2);
+      _mm_store_si128(dst_cacheline + 2, temp3);
+      _mm_store_si128(dst_cacheline + 3, temp4);
+
+      d += 64;
+      s += 64;
+      len -= 64;
+   }
+
+   /* memcpy() the tail. */
+   if (len) {
+      memcpy(d, s, len);
+   }
+}