Add intel opts to Makefile and configure.ac

Makefile.am

@@ -83,6 +83,11 @@ libpng@PNGLIB_MAJOR@@PNGLIB_MINOR@_la_SOURCES += arm/arm_init.c\
 	arm/filter_neon.S arm/filter_neon_intrinsics.c
 endif
 
+if PNG_INTEL_SSE
+libpng@PNGLIB_MAJOR@@PNGLIB_MINOR@_la_SOURCES += intel/intel_init.c\
+    intel/filter_sse2_intrinsics.c
+endif
+
 nodist_libpng@PNGLIB_MAJOR@@PNGLIB_MINOR@_la_SOURCES = pnglibconf.h
 
 libpng@PNGLIB_MAJOR@@PNGLIB_MINOR@_la_LDFLAGS = -no-undefined -export-dynamic \

configure.ac

@@ -314,6 +314,41 @@ AM_CONDITIONAL([PNG_ARM_NEON],
       *)    test "$enable_arm_neon" != '';;
     esac])
 
+# INTEL
+# ===
+#
+# INTEL SSE (SIMD) support.
+
+AC_ARG_ENABLE([intel-sse],
+   AS_HELP_STRING([[[--enable-intel-sse]]],
+      [Enable Intel SSE optimizations: =no/off, yes/on:]
+      [no/off: disable the optimizations;]
+      [yes/on: enable the optimizations.]
+      [If not specified: determined by the compiler.]),
+   [case "$enableval" in
+      no|off)
+         # disable the default enabling:
+         AC_DEFINE([PNG_INTEL_SSE_OPT], [0],
+                   [Disable Intel SSE optimizations])
+         # Prevent inclusion of the assembler files below:
+         enable_intel_sse=no;;
+      yes|on)
+         AC_DEFINE([PNG_INTEL_SSE_OPT], [1],
+                   [Enable Intel SSE optimizations]);;
+      *)
+         AC_MSG_ERROR([--enable-intel-sse=${enable_intel_sse}: invalid value])
+   esac])
+
+# Add Intel specific files to all builds where the host_cpu is Intel ('x86*')
+# or where Intel optimizations were explicitly requested (this allows a
+# fallback if a future host CPU does not match 'x86*')
+AM_CONDITIONAL([PNG_INTEL_SSE],
+   [test "$enable_intel_sse" != 'no' &&
+    case "$host_cpu" in
+      x86*) :;;
+      *)    test "$enable_intel_sse" != '';;
+    esac])
+
 AC_MSG_NOTICE([[Extra options for compiler: $PNG_COPTS]])
 
 # Config files, substituting as above

intel/filter_sse2_intrinsics.c

@@ -12,7 +12,7 @@
 
 #ifdef PNG_READ_SUPPORTED
 
-#if PNG_INTEL_SSE_OPT > 0
+#if PNG_INTEL_SSE_IMPLEMENTATION > 0
 
 #include <immintrin.h>
 
@@ -144,7 +144,7 @@ void png_read_filter_row_avg4_sse2(png_row_infop row_info, png_bytep row,
 
 // Returns |x| for 16-bit lanes.
 static __m128i abs_i16(__m128i x) {
-#if PNG_INTEL_SSE_OPT >= 2
+#if PNG_INTEL_SSE_IMPLEMENTATION >= 2
    return _mm_abs_epi16(x);
 #else
    // Read this all as, return x<0 ? -x : x.
@@ -162,7 +162,7 @@ static __m128i abs_i16(__m128i x) {
 
 // Bytewise c ? t : e.
 static __m128i if_then_else(__m128i c, __m128i t, __m128i e) {
-#if PNG_INTEL_SSE_OPT >= 3
+#if PNG_INTEL_SSE_IMPLEMENTATION >= 3
    return _mm_blendv_epi8(e,t,c);
 #else
    return _mm_or_si128(_mm_and_si128(c, t), _mm_andnot_si128(c, e));
@@ -281,5 +281,5 @@ void png_read_filter_row_paeth4_sse2(png_row_infop row_info, png_bytep row,
    }
 }
 
-#endif /* PNG_INTEL_SSE_OPT > 0 */
+#endif /* PNG_INTEL_SSE_IMPLEMENTATION > 0 */
 #endif /* READ */

intel/intel_init.c

@@ -11,7 +11,7 @@
 #include "../pngpriv.h"
 
 #ifdef PNG_READ_SUPPORTED
-#if PNG_INTEL_SSE_OPT > 0
+#if PNG_INTEL_SSE_IMPLEMENTATION > 0
 
 void
 png_init_filter_functions_sse2(png_structp pp, unsigned int bpp)
@@ -41,5 +41,5 @@ png_init_filter_functions_sse2(png_structp pp, unsigned int bpp)
    // No need optimize PNG_FILTER_VALUE_UP.  The compiler should autovectorize.
 }
 
-#endif /* PNG_INTEL_SSE_OPT > 0 */
+#endif /* PNG_INTEL_SSE_IMPLEMENTATION > 0 */
 #endif /* PNG_READ_SUPPORTED */

pngpriv.h

@@ -183,22 +183,34 @@
 #endif /* PNG_ARM_NEON_OPT > 0 */
 
 #ifndef PNG_INTEL_SSE_OPT
-#   if defined(__SSE4_1__) || defined(__AVX__)
+#   if defined(__SSE4_1__) || defined(__AVX__) || defined(__SSSE3__) || \
-       /* We are not actually using AVX, but checking for AVX is the best
+    defined(__SSE2__) || defined(_M_X64) || defined(_M_AMD64) || \
-          way we can detect SSE4.1 and SSSE3 on MSVC.
-       */
-#      define PNG_INTEL_SSE_OPT 3
-#   elif defined(__SSSE3__)
-#      define PNG_INTEL_SSE_OPT 2
-#   elif defined(__SSE2__) || defined(_M_X64) || defined(_M_AMD64) || \
     (defined(_M_IX86_FP) && _M_IX86_FP >= 2)
 #      define PNG_INTEL_SSE_OPT 1
 #   endif
 #endif
 
 #if PNG_INTEL_SSE_OPT > 0
+#   ifndef PNG_INTEL_SSE_IMPLEMENTATION
+#      if defined(__SSE4_1__) || defined(__AVX__)
+          /* We are not actually using AVX, but checking for AVX is the best
+             way we can detect SSE4.1 and SSSE3 on MSVC.
+          */
+#         define PNG_INTEL_SSE_IMPLEMENTATION 3
+#      elif defined(__SSSE3__)
+#         define PNG_INTEL_SSE_IMPLEMENTATION 2
+#      elif defined(__SSE2__) || defined(_M_X64) || defined(_M_AMD64) || \
+       (defined(_M_IX86_FP) && _M_IX86_FP >= 2)
+#         define PNG_INTEL_SSE_IMPLEMENTATION 1
+#      else
+#         define PNG_INTEL_SSE_IMPLEMENTATION 0
+#      endif
+#   endif
+
+#   if PNG_INTEL_SSE_IMPLEMENTATION > 0
 #      define PNG_FILTER_OPTIMIZATIONS png_init_filter_functions_sse2
 #   endif
+#endif
 
 /* Is this a build of a DLL where compilation of the object modules requires
  * different preprocessor settings to those required for a simple library?  If