[babl/sse-conversions-2013: 8/8] SSE Float, Int16: draft 3

[Daniel Sabo <daniels src gnome org>]

commit 67e0b310d407f5ee810a6b088e96edf081c48071
Author: Daniel Sabo <DanielSabo gmail com>
Date:   Sat Mar 30 18:29:51 2013 -0700

    SSE Float, Int16: draft 3
    
    Check for correct defines and include the SSE2 headers.
    
    Switch to by pixel conversion rather than planar, the by
    pixel version is  at least competitive with planar on all
    the systems I tested and noticeably faster on newer CPUs.
    
    I disabled the checks for GCC version because intrinsic
    based code should work on MSVC, however I have not
    actually tested that theory.

 extensions/sse-float.c |  125 +++++++++++++++++++++--------------------------
 extensions/sse-int16.c |   10 ++--
 2 files changed, 61 insertions(+), 74 deletions(-)
---
diff --git a/extensions/sse-float.c b/extensions/sse-float.c
index 1e486fd..bc3c42e 100644
--- a/extensions/sse-float.c
+++ b/extensions/sse-float.c
@@ -19,9 +19,10 @@
 
 #include "config.h"
 
-#if defined(__GNUC__) && (__GNUC__ >= 4) && defined(USE_SSE) && defined(USE_MMX)
+#if defined(USE_SSE2)
 
-#include <xmmintrin.h>
+/* SSE 2 */
+#include <emmintrin.h>
 
 #include <stdint.h>
 #include <stdlib.h>
@@ -40,41 +41,34 @@ conv_rgbaF_linear_rgbAF_linear (const float *src, float *dst, long samples)
 
   if (((uintptr_t)src % 16) + ((uintptr_t)dst % 16) == 0)
     {
-      const long    n = (samples / 4) * 4;
+      const long    n = (samples / 2) * 2;
       const __v4sf *s = (const __v4sf*) src;
             __v4sf *d = (__v4sf*)dst;
 
-      for ( ; i < n; i += 4)
+      for ( ; i < n; i += 2)
         {
-          const __v4sf s0 = *s++;
-          const __v4sf s1 = *s++;
-          const __v4sf s2 = *s++;
-          const __v4sf s3 = *s++;
-
-          /* Shuffle the pixels into a planar layout */
-          const __v4sf rg01 = _mm_unpacklo_ps (s0, s1);
-          const __v4sf ba01 = _mm_unpackhi_ps (s0, s1);
-          const __v4sf rg23 = _mm_unpacklo_ps (s2, s3);
-          const __v4sf ba23 = _mm_unpackhi_ps (s2, s3);
-
-          const __v4sf r0213 = _mm_unpacklo_ps (rg01, rg23);
-          const __v4sf g0213 = _mm_unpackhi_ps (rg01, rg23);
-          const __v4sf b0213 = _mm_unpacklo_ps (ba01, ba23);
-          const __v4sf a0213 = _mm_unpackhi_ps (ba01, ba23);
-
-          const __v4sf R0213 = r0213 * a0213;
-          const __v4sf G0213 = g0213 * a0213;
-          const __v4sf B0213 = b0213 * a0213;
-
-          const __v4sf RB02 = _mm_unpacklo_ps (R0213, B0213);
-          const __v4sf RB13 = _mm_unpackhi_ps (R0213, B0213);
-          const __v4sf Ga02 = _mm_unpacklo_ps (G0213, a0213);
-          const __v4sf Ga13 = _mm_unpackhi_ps (G0213, a0213);
-
-          *d++ = _mm_unpacklo_ps (RB02, Ga02);
-          *d++ = _mm_unpacklo_ps (RB13, Ga13);
-          *d++ = _mm_unpackhi_ps (RB02, Ga02);
-          *d++ = _mm_unpackhi_ps (RB13, Ga13);
+          __v4sf rbaa0, rbaa1;
+        
+          __v4sf rgba0 = *s++;
+          __v4sf rgba1 = *s++;
+
+          /* Expand alpha */
+          __v4sf aaaa0 = (__v4sf)_mm_shuffle_epi32((__m128i)rgba0, _MM_SHUFFLE(3, 3, 3, 3));
+          __v4sf aaaa1 = (__v4sf)_mm_shuffle_epi32((__m128i)rgba1, _MM_SHUFFLE(3, 3, 3, 3));
+          
+          /* Premultiply */
+          rgba0 = rgba0 * aaaa0;
+          rgba1 = rgba1 * aaaa1;
+          
+          /* Shuffle the original alpha value back in */
+          rbaa0 = _mm_shuffle_ps(rgba0, aaaa0, _MM_SHUFFLE(0, 0, 2, 0));
+          rbaa1 = _mm_shuffle_ps(rgba1, aaaa1, _MM_SHUFFLE(0, 0, 2, 0));
+          
+          rgba0 = _mm_shuffle_ps(rgba0, rbaa0, _MM_SHUFFLE(2, 1, 1, 0));
+          rgba1 = _mm_shuffle_ps(rgba1, rbaa1, _MM_SHUFFLE(2, 1, 1, 0));
+          
+          *d++ = rgba0;
+          *d++ = rgba1;
         }
       _mm_empty ();
     }
@@ -105,41 +99,33 @@ conv_rgbAF_linear_rgbaF_linear (const float *src, float *dst, long samples)
 
   if (((uintptr_t)src % 16) + ((uintptr_t)dst % 16) == 0)
     {
-      const long    n = (samples / 4) * 4;
+      const long    n = (samples / 2) * 2;
       const __v4sf *s = (const __v4sf*) src;
             __v4sf *d = (__v4sf*)dst;
 
-      for ( ; i < n; i += 4)
+      for ( ; i < n; i += 2)
         {
-          const __v4sf s0 = *s++;
-          const __v4sf s1 = *s++;
-          const __v4sf s2 = *s++;
-          const __v4sf s3 = *s++;
-
-          /* Shuffle the pixels into a planar layout */
-          const __v4sf rg01 = _mm_unpacklo_ps (s0, s1);
-          const __v4sf ba01 = _mm_unpackhi_ps (s0, s1);
-          const __v4sf rg23 = _mm_unpacklo_ps (s2, s3);
-          const __v4sf ba23 = _mm_unpackhi_ps (s2, s3);
-
-          const __v4sf r0213 = _mm_unpacklo_ps (rg01, rg23);
-          const __v4sf g0213 = _mm_unpackhi_ps (rg01, rg23);
-          const __v4sf b0213 = _mm_unpacklo_ps (ba01, ba23);
-          const __v4sf a0213 = _mm_unpackhi_ps (ba01, ba23);
-
-          const __v4sf R0213 = r0213 / a0213;
-          const __v4sf G0213 = g0213 / a0213;
-          const __v4sf B0213 = b0213 / a0213;
-
-          const __v4sf RB02 = _mm_unpacklo_ps (R0213, B0213);
-          const __v4sf RB13 = _mm_unpackhi_ps (R0213, B0213);
-          const __v4sf Ga02 = _mm_unpacklo_ps (G0213, a0213);
-          const __v4sf Ga13 = _mm_unpackhi_ps (G0213, a0213);
-
-          *d++ = _mm_unpacklo_ps (RB02, Ga02);
-          *d++ = _mm_unpacklo_ps (RB13, Ga13);
-          *d++ = _mm_unpackhi_ps (RB02, Ga02);
-          *d++ = _mm_unpackhi_ps (RB13, Ga13);
+          __v4sf rbaa0, rbaa1;
+          
+          __v4sf rgba0 = *s++;
+          __v4sf rgba1 = *s++;
+
+          /* Expand alpha */
+          __v4sf aaaa0 = (__v4sf)_mm_shuffle_epi32((__m128i)rgba0, _MM_SHUFFLE(3, 3, 3, 3));
+          __v4sf aaaa1 = (__v4sf)_mm_shuffle_epi32((__m128i)rgba1, _MM_SHUFFLE(3, 3, 3, 3));
+          
+          /* Premultiply */
+          rgba0 = rgba0 / aaaa0;
+          rgba1 = rgba1 / aaaa1;
+          
+          /* Shuffle the original alpha value back in */
+          rbaa0 = _mm_shuffle_ps(rgba0, aaaa0, _MM_SHUFFLE(0, 0, 2, 0));
+          rbaa1 = _mm_shuffle_ps(rgba1, aaaa1, _MM_SHUFFLE(0, 0, 2, 0));
+          rgba0 = _mm_shuffle_ps(rgba0, rbaa0, _MM_SHUFFLE(2, 1, 1, 0));
+          rgba1 = _mm_shuffle_ps(rgba1, rbaa1, _MM_SHUFFLE(2, 1, 1, 0));
+          
+          *d++ = rgba0;
+          *d++ = rgba1;
         }
       _mm_empty ();
     }
@@ -163,7 +149,7 @@ conv_rgbAF_linear_rgbaF_linear (const float *src, float *dst, long samples)
   return samples;
 }
 
-#endif /* defined(__GNUC__) && (__GNUC__ >= 4) && defined(USE_SSE) && defined(USE_MMX) */
+#endif /* defined(USE_SSE2) */
 
 #define o(src, dst) \
   babl_conversion_new (src, dst, "linear", conv_ ## src ## _ ## dst, NULL)
@@ -173,7 +159,7 @@ int init (void);
 int
 init (void)
 {
-#if defined(__GNUC__) && (__GNUC__ >= 4) && defined(USE_SSE) && defined(USE_MMX)
+#if defined(USE_SSE2)
 
   const Babl *rgbaF_linear = babl_format_new (
     babl_model ("RGBA"),
@@ -192,14 +178,15 @@ init (void)
     babl_component ("A"),
     NULL);
 
-  if ((babl_cpu_accel_get_support () & BABL_CPU_ACCEL_X86_MMX) &&
-      (babl_cpu_accel_get_support () & BABL_CPU_ACCEL_X86_SSE))
+  if ((babl_cpu_accel_get_support () & BABL_CPU_ACCEL_X86_SSE) &&
+      (babl_cpu_accel_get_support () & BABL_CPU_ACCEL_X86_SSE2))
+      
     {
       o (rgbaF_linear, rgbAF_linear);
       o (rgbAF_linear, rgbaF_linear);
     }
 
-#endif
+#endif /* defined(USE_SSE2) */
 
   return 0;
 }
diff --git a/extensions/sse-int16.c b/extensions/sse-int16.c
index 4118817..0876ea3 100644
--- a/extensions/sse-int16.c
+++ b/extensions/sse-int16.c
@@ -19,7 +19,7 @@
 
 #include "config.h"
 
-#if defined(__GNUC__) && (__GNUC__ >= 4) && defined(USE_SSE) && defined(USE_MMX)
+#if defined(USE_SSE2)
 
 /* SSE 2 */
 #include <emmintrin.h>
@@ -94,7 +94,7 @@ conv_rgba16_linear_rgbAF_linear (const uint16_t *src, float *dst, long samples)
           const __m128  u0 = _mm_cvtepi32_ps (t0);
           const __m128  u1 = _mm_cvtepi32_ps (t1);
 
-          /* Multiply by 1 / 2^16 */
+          /* Multiply by 1 / 65535 */
           __v4sf rgba0 = u0 * u16_float;
           __v4sf rgba1 = u1 * u16_float;
           
@@ -135,7 +135,7 @@ conv_rgba16_linear_rgbAF_linear (const uint16_t *src, float *dst, long samples)
   return samples;
 }
 
-#endif
+#endif /* defined(USE_SSE2) */
 
 #define o(src, dst) \
   babl_conversion_new (src, dst, "linear", conv_ ## src ## _ ## dst, NULL)
@@ -145,7 +145,7 @@ int init (void);
 int
 init (void)
 {
-#if defined(__GNUC__) && (__GNUC__ >= 4) && defined(USE_SSE) && defined(USE_MMX)
+#if defined(USE_SSE2)
 
   const Babl *rgbaF_linear = babl_format_new (
     babl_model ("RGBA"),
@@ -179,7 +179,7 @@ init (void)
       o (rgba16_linear, rgbAF_linear);
     }
 
-#endif
+#endif /* defined(USE_SSE2) */
 
   return 0;
 }