stub for SSE4.1 support.

Change-Id: I0c845a98d2871cc8907ff7b914bab7747a92c7ed

Android.mk

@@ -48,6 +48,7 @@ dsp_dec_srcs := \
     src/dsp/dec_mips_dsp_r2.c \
     src/dsp/dec_neon.$(NEON) \
     src/dsp/dec_sse2.c \
+    src/dsp/dec_sse41.c \
     src/dsp/filters.c \
     src/dsp/filters_mips_dsp_r2.c \
     src/dsp/filters_sse2.c \

Makefile.vc

@@ -195,6 +195,7 @@ DSP_DEC_OBJS = \
     $(DIROBJ)\dsp\dec_mips_dsp_r2.obj \
     $(DIROBJ)\dsp\dec_neon.obj \
     $(DIROBJ)\dsp\dec_sse2.obj \
+    $(DIROBJ)\dsp\dec_sse41.obj \
     $(DIROBJ)\dsp\filters.obj \
     $(DIROBJ)\dsp\filters_mips_dsp_r2.obj \
     $(DIROBJ)\dsp\filters_sse2.obj \

configure.ac

@@ -97,6 +97,17 @@ AS_IF([test -n "$AVX2_FLAGS"], [
   CFLAGS=$SAVED_CFLAGS])
 AC_SUBST([AVX2_FLAGS])
 
+TEST_AND_ADD_CFLAGS([SSE41_FLAGS], [-msse4.1])
+AS_IF([test -n "$SSE41_FLAGS"], [
+  SAVED_CFLAGS=$CFLAGS
+  CFLAGS="$CFLAGS $SSE41_FLAGS"
+  AC_CHECK_HEADER([smmintrin.h],
+                  [AC_DEFINE(WEBP_HAVE_SSE41, [1],
+                   [Set to 1 if SSE4.1 is supported])],
+                  [SSE41_FLAGS=""])
+  CFLAGS=$SAVED_CFLAGS])
+AC_SUBST([SSE41_FLAGS])
+
 TEST_AND_ADD_CFLAGS([SSE2_FLAGS], [-msse2])
 AS_IF([test -n "$SSE2_FLAGS"], [
   SAVED_CFLAGS=$CFLAGS

makefile.unix

@@ -119,6 +119,7 @@ DSP_DEC_OBJS = \
     src/dsp/dec_mips_dsp_r2.o \
     src/dsp/dec_neon.o \
     src/dsp/dec_sse2.o \
+    src/dsp/dec_sse41.o \
     src/dsp/filters.o \
     src/dsp/filters_mips_dsp_r2.o \
     src/dsp/filters_sse2.o \

src/dsp/Makefile.am

@@ -1,5 +1,6 @@
 noinst_LTLIBRARIES = libwebpdsp.la libwebpdsp_avx2.la
 noinst_LTLIBRARIES += libwebpdsp_sse2.la libwebpdspdecode_sse2.la
+noinst_LTLIBRARIES += libwebpdspdecode_sse41.la
 
 if BUILD_LIBWEBPDECODER
   noinst_LTLIBRARIES += libwebpdspdecode.la
@@ -54,6 +55,11 @@ libwebpdsp_avx2_la_SOURCES += enc_avx2.c
 libwebpdsp_avx2_la_CPPFLAGS = $(libwebpdsp_la_CPPFLAGS)
 libwebpdsp_avx2_la_CFLAGS = $(AM_CFLAGS) $(AVX2_FLAGS)
 
+libwebpdspdecode_sse41_la_SOURCES =
+libwebpdspdecode_sse41_la_SOURCES += dec_sse41.c
+libwebpdspdecode_sse41_la_CPPFLAGS = $(libwebpdsp_la_CPPFLAGS)
+libwebpdspdecode_sse41_la_CFLAGS = $(AM_CFLAGS) $(SSE41_FLAGS)
+
 libwebpdspdecode_sse2_la_SOURCES =
 libwebpdspdecode_sse2_la_SOURCES += alpha_processing_sse2.c
 libwebpdspdecode_sse2_la_SOURCES += dec_sse2.c
@@ -81,12 +87,16 @@ noinst_HEADERS += ../webp/decode.h
 
 libwebpdsp_la_CPPFLAGS = $(USE_EXPERIMENTAL_CODE) $(USE_SWAP_16BIT_CSP)
 libwebpdsp_la_LDFLAGS = -lm
-libwebpdsp_la_LIBADD = libwebpdsp_avx2.la libwebpdsp_sse2.la
+libwebpdsp_la_LIBADD =
+libwebpdsp_la_LIBADD += libwebpdsp_avx2.la libwebpdsp_sse2.la
+libwebpdsp_la_LIBADD += libwebpdspdecode_sse41.la
 
 if BUILD_LIBWEBPDECODER
   libwebpdspdecode_la_SOURCES = $(COMMON_SOURCES)
 
   libwebpdspdecode_la_CPPFLAGS = $(libwebpdsp_la_CPPFLAGS)
   libwebpdspdecode_la_LDFLAGS = $(libwebpdsp_la_LDFLAGS)
-  libwebpdspdecode_la_LIBADD = libwebpdspdecode_sse2.la
+  libwebpdspdecode_la_LIBADD =
+  libwebpdspdecode_la_LIBADD += libwebpdspdecode_sse2.la
+  libwebpdspdecode_la_LIBADD += libwebpdspdecode_sse41.la
 endif

src/dsp/dec.c

@@ -674,6 +674,7 @@ VP8SimpleFilterFunc VP8SimpleVFilter16i;
 VP8SimpleFilterFunc VP8SimpleHFilter16i;
 
 extern void VP8DspInitSSE2(void);
+extern void VP8DspInitSSE41(void);
 extern void VP8DspInitNEON(void);
 extern void VP8DspInitMIPS32(void);
 extern void VP8DspInitMIPSdspR2(void);
@@ -738,6 +739,11 @@ WEBP_TSAN_IGNORE_FUNCTION void VP8DspInit(void) {
 #if defined(WEBP_USE_SSE2)
     if (VP8GetCPUInfo(kSSE2)) {
       VP8DspInitSSE2();
+#if defined(WEBP_USE_SSE41)
+      if (VP8GetCPUInfo(kSSE4_1)) {
+        VP8DspInitSSE41();
+      }
+#endif
     }
 #endif
 #if defined(WEBP_USE_NEON)

src/dsp/dec_sse41.c

@@ -0,0 +1,38 @@
+// 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.
+// -----------------------------------------------------------------------------
+//
+// SSE4 version of some decoding functions.
+//
+// Author: Skal (pascal.massimino@gmail.com)
+
+#include "./dsp.h"
+
+#if defined(WEBP_USE_SSE41)
+
+#include <smmintrin.h>
+#include "../dec/vp8i.h"
+
+static void HE16(uint8_t* dst) {     // horizontal
+  int j;
+  const __m128i kShuffle3 = _mm_set1_epi8(3);
+  for (j = 16; j > 0; --j) {
+    const __m128i in = _mm_cvtsi32_si128(*(int*)(dst - 4));
+    const __m128i values = _mm_shuffle_epi8(in, kShuffle3);
+    _mm_storeu_si128((__m128i*)dst, values);
+    dst += BPS;
+  }
+}
+#endif   // WEBP_USE_SSE41
+
+extern void VP8DspInitSSE41(void);
+WEBP_TSAN_IGNORE_FUNCTION void VP8DspInitSSE41(void) {
+#if defined(WEBP_USE_SSE41)
+  VP8PredLuma16[3] = HE16;
+#endif   // WEBP_USE_SSE41
+}

src/dsp/dsp.h

@@ -52,6 +52,11 @@ extern "C" {
 #define WEBP_MSC_SSE2  // Visual C++ SSE2 targets
 #endif
 
+#if defined(_MSC_VER) && _MSC_VER >= 1500 && \
+    (defined(_M_X64) || defined(_M_IX86))
+#define WEBP_MSC_SSE41  // Visual C++ SSE4.1 targets
+#endif
+
 // WEBP_HAVE_* are used to indicate the presence of the instruction set in dsp
 // files without intrinsics, allowing the corresponding Init() to be called.
 // Files containing intrinsics will need to be built targeting the instruction
@@ -60,6 +65,10 @@ extern "C" {
 #define WEBP_USE_SSE2
 #endif
 
+#if defined(__SSE4_1__) || defined(WEBP_MSC_SSE41) || defined(WEBP_HAVE_SSE41)
+#define WEBP_USE_SSE41
+#endif
+
 #if defined(__AVX2__) || defined(WEBP_HAVE_AVX2)
 #define WEBP_USE_AVX2
 #endif