summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorIlya Kurdyukov <jpegqs@gmail.com>2021-02-18 12:36:31 +0100
committerSkal <skal@google.com>2021-02-18 12:42:39 +0100
commita09a647241a2ea17eda774a831ae07a1f4d461b1 (patch)
tree12284325ca3d4dd773de598b28a8d3b496be39f2
parent401da22bd6ad52586c4b4f2fc1d26804a78ccbbe (diff)
downloadlibwebp-a09a647241a2ea17eda774a831ae07a1f4d461b1.tar.gz
SSE4.1 version of TransformColorInverse
Change-Id: I6ba5cb35917eef7a52152c4924eca205b4af7220
-rw-r--r--Android.mk1
-rw-r--r--Makefile.vc1
-rw-r--r--build.gradle1
-rw-r--r--makefile.unix1
-rw-r--r--src/dsp/Makefile.am1
-rw-r--r--src/dsp/lossless.c6
-rw-r--r--src/dsp/lossless_sse41.c69
7 files changed, 80 insertions, 0 deletions
diff --git a/Android.mk b/Android.mk
index 8f0cb756..4ae099c3 100644
--- a/Android.mk
+++ b/Android.mk
@@ -74,6 +74,7 @@ dsp_dec_srcs := \
src/dsp/lossless_msa.c \
src/dsp/lossless_neon.$(NEON) \
src/dsp/lossless_sse2.c \
+ src/dsp/lossless_sse41.c \
src/dsp/rescaler.c \
src/dsp/rescaler_mips32.c \
src/dsp/rescaler_mips_dsp_r2.c \
diff --git a/Makefile.vc b/Makefile.vc
index feb62031..b9d631d2 100644
--- a/Makefile.vc
+++ b/Makefile.vc
@@ -215,6 +215,7 @@ DSP_DEC_OBJS = \
$(DIROBJ)\dsp\lossless_msa.obj \
$(DIROBJ)\dsp\lossless_neon.obj \
$(DIROBJ)\dsp\lossless_sse2.obj \
+ $(DIROBJ)\dsp\lossless_sse41.obj \
$(DIROBJ)\dsp\rescaler.obj \
$(DIROBJ)\dsp\rescaler_mips32.obj \
$(DIROBJ)\dsp\rescaler_mips_dsp_r2.obj \
diff --git a/build.gradle b/build.gradle
index e587bb6d..bbd9bcf0 100644
--- a/build.gradle
+++ b/build.gradle
@@ -141,6 +141,7 @@ model {
include "lossless_msa.c"
include "lossless_neon.$NEON"
include "lossless_sse2.c"
+ include "lossless_sse41.c"
include "rescaler.c"
include "rescaler_mips32.c"
include "rescaler_mips_dsp_r2.c"
diff --git a/makefile.unix b/makefile.unix
index 769c5494..aa7f3a29 100644
--- a/makefile.unix
+++ b/makefile.unix
@@ -166,6 +166,7 @@ DSP_DEC_OBJS = \
src/dsp/lossless_msa.o \
src/dsp/lossless_neon.o \
src/dsp/lossless_sse2.o \
+ src/dsp/lossless_sse41.o \
src/dsp/rescaler.o \
src/dsp/rescaler_mips32.o \
src/dsp/rescaler_mips_dsp_r2.o \
diff --git a/src/dsp/Makefile.am b/src/dsp/Makefile.am
index 9f67f5b3..766bb0a7 100644
--- a/src/dsp/Makefile.am
+++ b/src/dsp/Makefile.am
@@ -46,6 +46,7 @@ ENC_SOURCES += ssim.c
libwebpdspdecode_sse41_la_SOURCES =
libwebpdspdecode_sse41_la_SOURCES += alpha_processing_sse41.c
libwebpdspdecode_sse41_la_SOURCES += dec_sse41.c
+libwebpdspdecode_sse41_la_SOURCES += lossless_sse41.c
libwebpdspdecode_sse41_la_SOURCES += upsampling_sse41.c
libwebpdspdecode_sse41_la_SOURCES += yuv_sse41.c
libwebpdspdecode_sse41_la_CPPFLAGS = $(libwebpdsp_la_CPPFLAGS)
diff --git a/src/dsp/lossless.c b/src/dsp/lossless.c
index 46b220e2..ca8289fb 100644
--- a/src/dsp/lossless.c
+++ b/src/dsp/lossless.c
@@ -575,6 +575,7 @@ VP8LMapARGBFunc VP8LMapColor32b;
VP8LMapAlphaFunc VP8LMapColor8b;
extern void VP8LDspInitSSE2(void);
+extern void VP8LDspInitSSE41(void);
extern void VP8LDspInitNEON(void);
extern void VP8LDspInitMIPSdspR2(void);
extern void VP8LDspInitMSA(void);
@@ -624,6 +625,11 @@ WEBP_DSP_INIT_FUNC(VP8LDspInit) {
#if defined(WEBP_USE_SSE2)
if (VP8GetCPUInfo(kSSE2)) {
VP8LDspInitSSE2();
+#if defined(WEBP_USE_SSE41)
+ if (VP8GetCPUInfo(kSSE4_1)) {
+ VP8LDspInitSSE41();
+ }
+#endif
}
#endif
#if defined(WEBP_USE_MIPS_DSP_R2)
diff --git a/src/dsp/lossless_sse41.c b/src/dsp/lossless_sse41.c
new file mode 100644
index 00000000..739c3c74
--- /dev/null
+++ b/src/dsp/lossless_sse41.c
@@ -0,0 +1,69 @@
+// Copyright 2021 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.
+// -----------------------------------------------------------------------------
+//
+// SSE41 variant of methods for lossless decoder
+
+#include "src/dsp/dsp.h"
+
+#if defined(WEBP_USE_SSE41)
+
+#include "src/dsp/common_sse41.h"
+#include "src/dsp/lossless.h"
+#include "src/dsp/lossless_common.h"
+#include <assert.h>
+#include <emmintrin.h>
+
+//------------------------------------------------------------------------------
+// Color-space conversion functions
+
+static void TransformColorInverse_SSE41(const VP8LMultipliers* const m,
+ const uint32_t* const src,
+ int num_pixels, uint32_t* dst) {
+#define CST(X) ((int32_t)((int8_t)m->X) << 3)
+ const __m128i mults_rb = _mm_set1_epi32(CST(green_to_red_) << 16 |
+ (CST(green_to_blue_) & 0xffff));
+ const __m128i mults_b2 = _mm_set1_epi32(CST(red_to_blue_));
+#undef CST
+ const __m128i mask_ag = _mm_set1_epi32(0xff00ff00);
+ const __m128i perm1 = _mm_setr_epi8(-1, 1, -1, 1, -1, 5, -1, 5,
+ -1, 9, -1, 9, -1, 13, -1, 13);
+ const __m128i perm2 = _mm_setr_epi8(-1, 2, -1, -1, -1, 6, -1, -1,
+ -1, 10, -1, -1, -1, 14, -1, -1);
+ int i;
+ for (i = 0; i + 4 <= num_pixels; i += 4) {
+ const __m128i A = _mm_loadu_si128((const __m128i*)(src + i));
+ const __m128i B = _mm_shuffle_epi8(A, perm1); // argb -> g0g0
+ const __m128i C = _mm_mulhi_epi16(B, mults_rb);
+ const __m128i D = _mm_add_epi8(A, C);
+ const __m128i E = _mm_shuffle_epi8(D, perm2);
+ const __m128i F = _mm_mulhi_epi16(E, mults_b2);
+ const __m128i G = _mm_add_epi8(D, F);
+ const __m128i out = _mm_blendv_epi8(G, A, mask_ag);
+ _mm_storeu_si128((__m128i*)&dst[i], out);
+ }
+ // Fall-back to C-version for left-overs.
+ if (i != num_pixels) {
+ VP8LTransformColorInverse_C(m, src + i, num_pixels - i, dst + i);
+ }
+}
+
+//------------------------------------------------------------------------------
+// Entry point
+
+extern void VP8LDspInitSSE41(void);
+
+WEBP_TSAN_IGNORE_FUNCTION void VP8LDspInitSSE41(void) {
+ VP8LTransformColorInverse = TransformColorInverse_SSE41;
+}
+
+#else // !WEBP_USE_SSE41
+
+WEBP_DSP_INIT_STUB(VP8LDspInitSSE41)
+
+#endif // WEBP_USE_SSE41