Add BGRA output format support to qcms

third_party/qcms/src/transform-sse1.c

 //  qcms
 //  Copyright (C) 2009 Mozilla Foundation
 //
 // Permission is hereby granted, free of charge, to any person obtaining
 // a copy of this software and associated documentation files (the "Software"),
 // to deal in the Software without restriction, including without limitation
 // the rights to use, copy, modify, merge, publish, distribute, sublicense,
 // and/or sell copies of the Software, and to permit persons to whom the Software
 // is furnished to do so, subject to the following conditions:
 //
@@ skipping 16 matching lines @@
 #define FLOATSCALE  (float)(PRECACHE_OUTPUT_SIZE)
 #define CLAMPMAXVAL ( ((float) (PRECACHE_OUTPUT_SIZE - 1)) / PRECACHE_OUTPUT_SIZE )
 static const ALIGN float floatScaleX4[4] =
     { FLOATSCALE, FLOATSCALE, FLOATSCALE, FLOATSCALE};
 static const ALIGN float clampMaxValueX4[4] =
     { CLAMPMAXVAL, CLAMPMAXVAL, CLAMPMAXVAL, CLAMPMAXVAL};
 
 void qcms_transform_data_rgb_out_lut_sse1(qcms_transform *transform,
                                           unsigned char *src,
                                           unsigned char *dest,
-                                          size_t length)
+                                          size_t length,
+                                          qcms_format_type output_format)
 {
     unsigned int i;
     float (*mat)[4] = transform->matrix;
     char input_back[32];
     /* Ensure we have a buffer that's 16 byte aligned regardless of the original
      * stack alignment. We can't use __attribute__((aligned(16))) or __declspec(align(32))
      * because they don't work on stack variables. gcc 4.4 does do the right thing
      * on x86 but that's too new for us right now. For more info: gcc bug #16660 */
     float const * input = (float*)(((uintptr_t)&input_back[16]) & ~0xf);
     /* share input and output locations to save having to keep the
@@ skipping 15 matching lines @@
     const __m128 mat1  = _mm_load_ps(mat[1]);
     const __m128 mat2  = _mm_load_ps(mat[2]);
 
     /* these values don't change, either */
     const __m128 max   = _mm_load_ps(clampMaxValueX4);
     const __m128 min   = _mm_setzero_ps();
     const __m128 scale = _mm_load_ps(floatScaleX4);
 
     /* working variables */
     __m128 vec_r, vec_g, vec_b, result;
+    const int r_out = output_format.r;
+    const int b_out = output_format.b;
 
     /* CYA */
     if (!length)
         return;
 
     /* one pixel is handled outside of the loop */
     length--;
 
     /* setup for transforming 1st pixel */
     vec_r = _mm_load_ss(&igtbl_r[src[0]]);
@@ skipping 26 matching lines @@
         result = _mm_movehl_ps(result, result);
         *((__m64 *)&output[2]) = _mm_cvtps_pi32(result) ;
 
         /* load for next loop while store completes */
         vec_r = _mm_load_ss(&igtbl_r[src[0]]);
         vec_g = _mm_load_ss(&igtbl_g[src[1]]);
         vec_b = _mm_load_ss(&igtbl_b[src[2]]);
         src += 3;
 
         /* use calc'd indices to output RGB values */
-        dest[0] = otdata_r[output[0]];
-        dest[1] = otdata_g[output[1]];
-        dest[2] = otdata_b[output[2]];
+        dest[r_out] = otdata_r[output[0]];
+        dest[1]     = otdata_g[output[1]];
+        dest[b_out] = otdata_b[output[2]];
         dest += 3;
     }
 
     /* handle final (maybe only) pixel */
 
     vec_r = _mm_shuffle_ps(vec_r, vec_r, 0);
     vec_g = _mm_shuffle_ps(vec_g, vec_g, 0);
     vec_b = _mm_shuffle_ps(vec_b, vec_b, 0);
 
     vec_r = _mm_mul_ps(vec_r, mat0);
     vec_g = _mm_mul_ps(vec_g, mat1);
     vec_b = _mm_mul_ps(vec_b, mat2);
 
     vec_r  = _mm_add_ps(vec_r, _mm_add_ps(vec_g, vec_b));
     vec_r  = _mm_max_ps(min, vec_r);
     vec_r  = _mm_min_ps(max, vec_r);
     result = _mm_mul_ps(vec_r, scale);
 
     *((__m64 *)&output[0]) = _mm_cvtps_pi32(result);
     result = _mm_movehl_ps(result, result);
     *((__m64 *)&output[2]) = _mm_cvtps_pi32(result);
 
-    dest[0] = otdata_r[output[0]];
-    dest[1] = otdata_g[output[1]];
-    dest[2] = otdata_b[output[2]];
+    dest[r_out] = otdata_r[output[0]];
+    dest[1]     = otdata_g[output[1]];
+    dest[b_out] = otdata_b[output[2]];
 
     _mm_empty();
 }
 
 void qcms_transform_data_rgba_out_lut_sse1(qcms_transform *transform,
                                            unsigned char *src,
                                            unsigned char *dest,
-                                           size_t length)
+                                           size_t length,
+                                           qcms_format_type output_format)
 {
     unsigned int i;
     float (*mat)[4] = transform->matrix;
     char input_back[32];
     /* Ensure we have a buffer that's 16 byte aligned regardless of the original
      * stack alignment. We can't use __attribute__((aligned(16))) or __declspec(align(32))
      * because they don't work on stack variables. gcc 4.4 does do the right thing
      * on x86 but that's too new for us right now. For more info: gcc bug #16660 */
     float const * input = (float*)(((uintptr_t)&input_back[16]) & ~0xf);
     /* share input and output locations to save having to keep the
@@ skipping 15 matching lines @@
     const __m128 mat1  = _mm_load_ps(mat[1]);
     const __m128 mat2  = _mm_load_ps(mat[2]);
 
     /* these values don't change, either */
     const __m128 max   = _mm_load_ps(clampMaxValueX4);
     const __m128 min   = _mm_setzero_ps();
     const __m128 scale = _mm_load_ps(floatScaleX4);
 
     /* working variables */
     __m128 vec_r, vec_g, vec_b, result;
+    const int r_out = output_format.r;
+    const int b_out = output_format.b;
     unsigned char alpha;
 
     /* CYA */
     if (!length)
         return;
 
     /* one pixel is handled outside of the loop */
     length--;
 
     /* setup for transforming 1st pixel */
@@ skipping 32 matching lines @@
         result = _mm_movehl_ps(result, result);
         *((__m64 *)&output[2]) = _mm_cvtps_pi32(result);
 
         /* load gamma values for next loop while store completes */
         vec_r = _mm_load_ss(&igtbl_r[src[0]]);
         vec_g = _mm_load_ss(&igtbl_g[src[1]]);
         vec_b = _mm_load_ss(&igtbl_b[src[2]]);
         src += 4;
 
         /* use calc'd indices to output RGB values */
-        dest[0] = otdata_r[output[0]];
-        dest[1] = otdata_g[output[1]];
-        dest[2] = otdata_b[output[2]];
+        dest[r_out] = otdata_r[output[0]];
+        dest[1]     = otdata_g[output[1]];
+        dest[b_out] = otdata_b[output[2]];
         dest += 4;
     }
 
     /* handle final (maybe only) pixel */
 
     vec_r = _mm_shuffle_ps(vec_r, vec_r, 0);
     vec_g = _mm_shuffle_ps(vec_g, vec_g, 0);
     vec_b = _mm_shuffle_ps(vec_b, vec_b, 0);
 
     vec_r = _mm_mul_ps(vec_r, mat0);
     vec_g = _mm_mul_ps(vec_g, mat1);
     vec_b = _mm_mul_ps(vec_b, mat2);
 
     dest[3] = alpha;
 
     vec_r  = _mm_add_ps(vec_r, _mm_add_ps(vec_g, vec_b));
     vec_r  = _mm_max_ps(min, vec_r);
     vec_r  = _mm_min_ps(max, vec_r);
     result = _mm_mul_ps(vec_r, scale);
 
     *((__m64 *)&output[0]) = _mm_cvtps_pi32(result);
     result = _mm_movehl_ps(result, result);
     *((__m64 *)&output[2]) = _mm_cvtps_pi32(result);
 
-    dest[0] = otdata_r[output[0]];
-    dest[1] = otdata_g[output[1]];
-    dest[2] = otdata_b[output[2]];
+    dest[r_out] = otdata_r[output[0]];
+    dest[1]     = otdata_g[output[1]];
+    dest[b_out] = otdata_b[output[2]];
 
     _mm_empty();
 }