summaryrefslogtreecommitdiffstats
path: root/media/libwebp/dsp/alpha_processing_sse41.c
diff options
context:
space:
mode:
authorwolfbeast <mcwerewolf@gmail.com>2017-07-24 13:19:25 +0200
committerwolfbeast <mcwerewolf@gmail.com>2018-02-03 09:57:42 +0100
commit2c1f833a092ef7af17f072749940bcb47da19aa6 (patch)
treee1904491a9c9a745d77df2225b679d921156affc /media/libwebp/dsp/alpha_processing_sse41.c
parentc0cfd6f89cd6b25cc8aaea1a1eb7bc004c88c206 (diff)
downloadUXP-2c1f833a092ef7af17f072749940bcb47da19aa6.tar
UXP-2c1f833a092ef7af17f072749940bcb47da19aa6.tar.gz
UXP-2c1f833a092ef7af17f072749940bcb47da19aa6.tar.lz
UXP-2c1f833a092ef7af17f072749940bcb47da19aa6.tar.xz
UXP-2c1f833a092ef7af17f072749940bcb47da19aa6.zip
Add libwebp 0.6.0 to the tree
Diffstat (limited to 'media/libwebp/dsp/alpha_processing_sse41.c')
-rw-r--r--media/libwebp/dsp/alpha_processing_sse41.c92
1 files changed, 92 insertions, 0 deletions
diff --git a/media/libwebp/dsp/alpha_processing_sse41.c b/media/libwebp/dsp/alpha_processing_sse41.c
new file mode 100644
index 000000000..986fde94e
--- /dev/null
+++ b/media/libwebp/dsp/alpha_processing_sse41.c
@@ -0,0 +1,92 @@
+// Copyright 2015 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING file in the root of the source
+// tree. An additional intellectual property rights grant can be found
+// in the file PATENTS. All contributing project authors may
+// be found in the AUTHORS file in the root of the source tree.
+// -----------------------------------------------------------------------------
+//
+// Utilities for processing transparent channel, SSE4.1 variant.
+//
+// Author: Skal (pascal.massimino@gmail.com)
+
+#include "./dsp.h"
+
+#if defined(WEBP_USE_SSE41)
+
+#include <smmintrin.h>
+
+//------------------------------------------------------------------------------
+
+static int ExtractAlpha(const uint8_t* argb, int argb_stride,
+ int width, int height,
+ uint8_t* alpha, int alpha_stride) {
+ // alpha_and stores an 'and' operation of all the alpha[] values. The final
+ // value is not 0xff if any of the alpha[] is not equal to 0xff.
+ uint32_t alpha_and = 0xff;
+ int i, j;
+ const __m128i all_0xff = _mm_set1_epi32(~0u);
+ __m128i all_alphas = all_0xff;
+
+ // We must be able to access 3 extra bytes after the last written byte
+ // 'src[4 * width - 4]', because we don't know if alpha is the first or the
+ // last byte of the quadruplet.
+ const int limit = (width - 1) & ~15;
+ const __m128i kCstAlpha0 = _mm_set_epi8(-1, -1, -1, -1, -1, -1, -1, -1,
+ -1, -1, -1, -1, 12, 8, 4, 0);
+ const __m128i kCstAlpha1 = _mm_set_epi8(-1, -1, -1, -1, -1, -1, -1, -1,
+ 12, 8, 4, 0, -1, -1, -1, -1);
+ const __m128i kCstAlpha2 = _mm_set_epi8(-1, -1, -1, -1, 12, 8, 4, 0,
+ -1, -1, -1, -1, -1, -1, -1, -1);
+ const __m128i kCstAlpha3 = _mm_set_epi8(12, 8, 4, 0, -1, -1, -1, -1,
+ -1, -1, -1, -1, -1, -1, -1, -1);
+ for (j = 0; j < height; ++j) {
+ const __m128i* src = (const __m128i*)argb;
+ for (i = 0; i < limit; i += 16) {
+ // load 64 argb bytes
+ const __m128i a0 = _mm_loadu_si128(src + 0);
+ const __m128i a1 = _mm_loadu_si128(src + 1);
+ const __m128i a2 = _mm_loadu_si128(src + 2);
+ const __m128i a3 = _mm_loadu_si128(src + 3);
+ const __m128i b0 = _mm_shuffle_epi8(a0, kCstAlpha0);
+ const __m128i b1 = _mm_shuffle_epi8(a1, kCstAlpha1);
+ const __m128i b2 = _mm_shuffle_epi8(a2, kCstAlpha2);
+ const __m128i b3 = _mm_shuffle_epi8(a3, kCstAlpha3);
+ const __m128i c0 = _mm_or_si128(b0, b1);
+ const __m128i c1 = _mm_or_si128(b2, b3);
+ const __m128i d0 = _mm_or_si128(c0, c1);
+ // store
+ _mm_storeu_si128((__m128i*)&alpha[i], d0);
+ // accumulate sixteen alpha 'and' in parallel
+ all_alphas = _mm_and_si128(all_alphas, d0);
+ src += 4;
+ }
+ for (; i < width; ++i) {
+ const uint32_t alpha_value = argb[4 * i];
+ alpha[i] = alpha_value;
+ alpha_and &= alpha_value;
+ }
+ argb += argb_stride;
+ alpha += alpha_stride;
+ }
+ // Combine the sixteen alpha 'and' into an 8-bit mask.
+ alpha_and |= 0xff00u; // pretend the upper bits [8..15] were tested ok.
+ alpha_and &= _mm_movemask_epi8(_mm_cmpeq_epi8(all_alphas, all_0xff));
+ return (alpha_and == 0xffffu);
+}
+
+//------------------------------------------------------------------------------
+// Entry point
+
+extern void WebPInitAlphaProcessingSSE41(void);
+
+WEBP_TSAN_IGNORE_FUNCTION void WebPInitAlphaProcessingSSE41(void) {
+ WebPExtractAlpha = ExtractAlpha;
+}
+
+#else // !WEBP_USE_SSE41
+
+WEBP_DSP_INIT_STUB(WebPInitAlphaProcessingSSE41)
+
+#endif // WEBP_USE_SSE41