util: Add support for other DXTn RGBA formats.

2 files changed, 298 insertions, 22 deletions

diff --git a/src/gallium/auxiliary/util/u_format_s3tc.c b/src/gallium/auxiliary/util/u_format_s3tc.c
index bb1d6d5ee07..268f0e8dfab 100644
--- a/src/gallium/auxiliary/util/u_format_s3tc.c
+++ b/src/gallium/auxiliary/util/u_format_s3tc.c
@@ -181,8 +181,6 @@ util_format_dxt5_rgba_fetch_float(float *dst, const uint8_t *src, unsigned i, un
 
 /*
  * Block decompression.
- *
- * FIXME
  */
 
 void
@@ -230,11 +228,43 @@ util_format_dxt1_rgba_unpack_8unorm(uint8_t *dst_row, unsigned dst_stride, const
 void
 util_format_dxt3_rgba_unpack_8unorm(uint8_t *dst_row, unsigned dst_stride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height)
 {
+   if (util_format_dxt3_rgba_fetch) {
+      unsigned x, y, i, j;
+      for(y = 0; y < height; y += 4) {
+         const uint8_t *src = src_row;
+         for(x = 0; x < width; x += 4) {
+            for(j = 0; j < 4; ++j) {
+               for(i = 0; i < 4; ++i) {
+                  uint8_t *dst = dst_row + (y + j)*dst_stride/sizeof(*dst_row) + (x + i)*4;
+                  util_format_dxt3_rgba_fetch(0, src, i, j, dst);
+               }
+            }
+            src += 16;
+         }
+         src_row += src_stride;
+      }
+   }
 }
 
 void
 util_format_dxt5_rgba_unpack_8unorm(uint8_t *dst_row, unsigned dst_stride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height)
 {
+   if (util_format_dxt5_rgba_fetch) {
+      unsigned x, y, i, j;
+      for(y = 0; y < height; y += 4) {
+         const uint8_t *src = src_row;
+         for(x = 0; x < width; x += 4) {
+            for(j = 0; j < 4; ++j) {
+               for(i = 0; i < 4; ++i) {
+                  uint8_t *dst = dst_row + (y + j)*dst_stride/sizeof(*dst_row) + (x + i)*4;
+                  util_format_dxt5_rgba_fetch(0, src, i, j, dst);
+               }
+            }
+            src += 16;
+         }
+         src_row += src_stride;
+      }
+   }
 }
 
 void
@@ -344,8 +374,6 @@ util_format_dxt5_rgba_unpack_float(float *dst_row, unsigned dst_stride, const ui
 
 /*
  * Block compression.
- *
- * FIXME
  */
 
 void
@@ -365,7 +393,7 @@ util_format_dxt1_rgb_pack_8unorm(uint8_t *dst_row, unsigned dst_stride, const ui
                   }
                }
             }
-            util_format_dxtn_pack(3, 4, 4, src, UTIL_FORMAT_DXT1_RGB, dst, dst_stride);
+            util_format_dxtn_pack(3, 4, 4, &tmp[0][0][0], UTIL_FORMAT_DXT1_RGB, dst, dst_stride);
             src += 4*4;
             dst += 8;
          }
@@ -378,36 +406,190 @@ util_format_dxt1_rgb_pack_8unorm(uint8_t *dst_row, unsigned dst_stride, const ui
 void
 util_format_dxt1_rgba_pack_8unorm(uint8_t *dst_row, unsigned dst_stride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height)
 {
+   if (util_format_dxtn_pack) {
+      unsigned x, y, i, j, k;
+      for(y = 0; y < height; y += 4) {
+         const uint8_t *src = src_row;
+         uint8_t *dst = dst_row;
+         for(x = 0; x < width; x += 4) {
+            uint8_t tmp[4][4][4];
+            for(j = 0; j < 4; ++j) {
+               for(i = 0; i < 4; ++i) {
+                  for(k = 0; k < 4; ++k) {
+                     tmp[j][i][k] = src[(y + j)*src_stride/sizeof(*src) + i*4 + k];
+                  }
+               }
+            }
+            util_format_dxtn_pack(4, 4, 4, &tmp[0][0][0], UTIL_FORMAT_DXT1_RGBA, dst, dst_stride);
+            src += 4*4;
+            dst += 8;
+         }
+         src_row += src_stride;
+         dst_row += 4*dst_stride/sizeof(*dst_row);
+      }
+   }
 }
 
 void
 util_format_dxt3_rgba_pack_8unorm(uint8_t *dst_row, unsigned dst_stride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height)
 {
+   if (util_format_dxtn_pack) {
+      unsigned x, y, i, j, k;
+      for(y = 0; y < height; y += 4) {
+         const uint8_t *src = src_row;
+         uint8_t *dst = dst_row;
+         for(x = 0; x < width; x += 4) {
+            uint8_t tmp[4][4][4];
+            for(j = 0; j < 4; ++j) {
+               for(i = 0; i < 4; ++i) {
+                  for(k = 0; k < 4; ++k) {
+                     tmp[j][i][k] = src[(y + j)*src_stride/sizeof(*src) + i*4 + k];
+                  }
+               }
+            }
+            util_format_dxtn_pack(4, 4, 4, &tmp[0][0][0], UTIL_FORMAT_DXT3_RGBA, dst, dst_stride);
+            src += 4*4;
+            dst += 16;
+         }
+         src_row += src_stride;
+         dst_row += 4*dst_stride/sizeof(*dst_row);
+      }
+   }
 }
 
 void
 util_format_dxt5_rgba_pack_8unorm(uint8_t *dst_row, unsigned dst_stride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height)
 {
+   if (util_format_dxtn_pack) {
+      unsigned x, y, i, j, k;
+      for(y = 0; y < height; y += 4) {
+         const uint8_t *src = src_row;
+         uint8_t *dst = dst_row;
+         for(x = 0; x < width; x += 4) {
+            uint8_t tmp[4][4][4];
+            for(j = 0; j < 4; ++j) {
+               for(i = 0; i < 4; ++i) {
+                  for(k = 0; k < 4; ++k) {
+                     tmp[j][i][k] = src[(y + j)*src_stride/sizeof(*src) + i*4 + k];
+                  }
+               }
+            }
+            util_format_dxtn_pack(4, 4, 4, &tmp[0][0][0], UTIL_FORMAT_DXT5_RGBA, dst, dst_stride);
+            src += 4*4;
+            dst += 16;
+         }
+         src_row += src_stride;
+         dst_row += 4*dst_stride/sizeof(*dst_row);
+      }
+   }
 }
 
 void
 util_format_dxt1_rgb_pack_float(uint8_t *dst_row, unsigned dst_stride, const float *src_row, unsigned src_stride, unsigned width, unsigned height)
 {
+   if (util_format_dxtn_pack) {
+      unsigned x, y, i, j, k;
+      for(y = 0; y < height; y += 4) {
+         const float *src = src_row;
+         uint8_t *dst = dst_row;
+         for(x = 0; x < width; x += 4) {
+            uint8_t tmp[4][4][3];
+            for(j = 0; j < 4; ++j) {
+               for(i = 0; i < 4; ++i) {
+                  for(k = 0; k < 3; ++k) {
+                     tmp[j][i][k] = float_to_ubyte(src[(y + j)*src_stride/sizeof(*src) + i*4 + k]);
+                  }
+               }
+            }
+            util_format_dxtn_pack(3, 4, 4, &tmp[0][0][0], UTIL_FORMAT_DXT1_RGB, dst, dst_stride);
+            src += 4*4;
+            dst += 8;
+         }
+         src_row += src_stride;
+         dst_row += 4*dst_stride/sizeof(*dst_row);
+      }
+   }
 }
 
 void
 util_format_dxt1_rgba_pack_float(uint8_t *dst_row, unsigned dst_stride, const float *src_row, unsigned src_stride, unsigned width, unsigned height)
 {
+   if (util_format_dxtn_pack) {
+      unsigned x, y, i, j, k;
+      for(y = 0; y < height; y += 4) {
+         const float *src = src_row;
+         uint8_t *dst = dst_row;
+         for(x = 0; x < width; x += 4) {
+            uint8_t tmp[4][4][4];
+            for(j = 0; j < 4; ++j) {
+               for(i = 0; i < 4; ++i) {
+                  for(k = 0; k < 4; ++k) {
+                     tmp[j][i][k] = float_to_ubyte(src[(y + j)*src_stride/sizeof(*src) + i*4 + k]);
+                  }
+               }
+            }
+            util_format_dxtn_pack(4, 4, 4, &tmp[0][0][0], UTIL_FORMAT_DXT1_RGBA, dst, dst_stride);
+            src += 4*4;
+            dst += 8;
+         }
+         src_row += src_stride;
+         dst_row += 4*dst_stride/sizeof(*dst_row);
+      }
+   }
 }
 
 void
 util_format_dxt3_rgba_pack_float(uint8_t *dst_row, unsigned dst_stride, const float *src_row, unsigned src_stride, unsigned width, unsigned height)
 {
+   if (util_format_dxtn_pack) {
+      unsigned x, y, i, j, k;
+      for(y = 0; y < height; y += 4) {
+         const float *src = src_row;
+         uint8_t *dst = dst_row;
+         for(x = 0; x < width; x += 4) {
+            uint8_t tmp[4][4][4];
+            for(j = 0; j < 4; ++j) {
+               for(i = 0; i < 4; ++i) {
+                  for(k = 0; k < 4; ++k) {
+                     tmp[j][i][k] = float_to_ubyte(src[(y + j)*src_stride/sizeof(*src) + i*4 + k]);
+                  }
+               }
+            }
+            util_format_dxtn_pack(4, 4, 4, &tmp[0][0][0], UTIL_FORMAT_DXT3_RGBA, dst, dst_stride);
+            src += 4*4;
+            dst += 16;
+         }
+         src_row += src_stride;
+         dst_row += 4*dst_stride/sizeof(*dst_row);
+      }
+   }
 }
 
 void
 util_format_dxt5_rgba_pack_float(uint8_t *dst_row, unsigned dst_stride, const float *src_row, unsigned src_stride, unsigned width, unsigned height)
 {
+   if (util_format_dxtn_pack) {
+      unsigned x, y, i, j, k;
+      for(y = 0; y < height; y += 4) {
+         const float *src = src_row;
+         uint8_t *dst = dst_row;
+         for(x = 0; x < width; x += 4) {
+            uint8_t tmp[4][4][4];
+            for(j = 0; j < 4; ++j) {
+               for(i = 0; i < 4; ++i) {
+                  for(k = 0; k < 4; ++k) {
+                     tmp[j][i][k] = float_to_ubyte(src[(y + j)*src_stride/sizeof(*src) + i*4 + k]);
+                  }
+               }
+            }
+            util_format_dxtn_pack(4, 4, 4, &tmp[0][0][0], UTIL_FORMAT_DXT5_RGBA, dst, dst_stride);
+            src += 4*4;
+            dst += 16;
+         }
+         src_row += src_stride;
+         dst_row += 4*dst_stride/sizeof(*dst_row);
+      }
+   }
 }
 
 
diff --git a/src/gallium/auxiliary/util/u_format_tests.c b/src/gallium/auxiliary/util/u_format_tests.c
index 816f4fd985f..0f81dcb79bc 100644
--- a/src/gallium/auxiliary/util/u_format_tests.c
+++ b/src/gallium/auxiliary/util/u_format_tests.c
@@ -205,7 +205,7 @@ util_format_test_cases[] =
     */
 
    /*
-    * TODO: Compressed formats
+    * Compressed formats
     */
 
    {
@@ -214,31 +214,125 @@ util_format_test_cases[] =
       PACKED_8x8(0xf2, 0xd7, 0xb0, 0x20, 0xae, 0x2c, 0x6f, 0x97),
       {
          {
-            0x99/255.0, 0xb0/255.0, 0x8e/255.0, 0xff/255.0,
-            0x5d/255.0, 0x62/255.0, 0x89/255.0, 0xff/255.0,
-            0x99/255.0, 0xb0/255.0, 0x8e/255.0, 0xff/255.0,
-            0x99/255.0, 0xb0/255.0, 0x8e/255.0, 0xff/255.0
+            {0x99/255.0, 0xb0/255.0, 0x8e/255.0, 0xff/255.0},
+            {0x5d/255.0, 0x62/255.0, 0x89/255.0, 0xff/255.0},
+            {0x99/255.0, 0xb0/255.0, 0x8e/255.0, 0xff/255.0},
+            {0x99/255.0, 0xb0/255.0, 0x8e/255.0, 0xff/255.0}
          },
          {
-            0xd6/255.0, 0xff/255.0, 0x94/255.0, 0xff/255.0,
-            0x5d/255.0, 0x62/255.0, 0x89/255.0, 0xff/255.0,
-            0x99/255.0, 0xb0/255.0, 0x8e/255.0, 0xff/255.0,
-            0xd6/255.0, 0xff/255.0, 0x94/255.0, 0xff/255.0
+            {0xd6/255.0, 0xff/255.0, 0x94/255.0, 0xff/255.0},
+            {0x5d/255.0, 0x62/255.0, 0x89/255.0, 0xff/255.0},
+            {0x99/255.0, 0xb0/255.0, 0x8e/255.0, 0xff/255.0},
+            {0xd6/255.0, 0xff/255.0, 0x94/255.0, 0xff/255.0}
          },
          {
-            0x5d/255.0, 0x62/255.0, 0x89/255.0, 0xff/255.0,
-            0x5d/255.0, 0x62/255.0, 0x89/255.0, 0xff/255.0,
-            0x99/255.0, 0xb0/255.0, 0x8e/255.0, 0xff/255.0,
-            0x21/255.0, 0x14/255.0, 0x84/255.0, 0xff/255.0
+            {0x5d/255.0, 0x62/255.0, 0x89/255.0, 0xff/255.0},
+            {0x5d/255.0, 0x62/255.0, 0x89/255.0, 0xff/255.0},
+            {0x99/255.0, 0xb0/255.0, 0x8e/255.0, 0xff/255.0},
+            {0x21/255.0, 0x14/255.0, 0x84/255.0, 0xff/255.0}
          },
          {
-            0x5d/255.0, 0x62/255.0, 0x89/255.0, 0xff/255.0,
-            0x21/255.0, 0x14/255.0, 0x84/255.0, 0xff/255.0,
-            0x21/255.0, 0x14/255.0, 0x84/255.0, 0xff/255.0,
-            0x99/255.0, 0xb0/255.0, 0x8e/255.0, 0xff/255.0
+            {0x5d/255.0, 0x62/255.0, 0x89/255.0, 0xff/255.0},
+            {0x21/255.0, 0x14/255.0, 0x84/255.0, 0xff/255.0},
+            {0x21/255.0, 0x14/255.0, 0x84/255.0, 0xff/255.0},
+            {0x99/255.0, 0xb0/255.0, 0x8e/255.0, 0xff/255.0}
          }
       }
    },
+   {
+      PIPE_FORMAT_DXT1_RGBA,
+      PACKED_8x8(0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff),
+      PACKED_8x8(0xff, 0x2f, 0xa4, 0x72, 0xeb, 0xb2, 0xbd, 0xbe),
+      {
+         {
+            {0x00/255.0, 0x00/255.0, 0x00/255.0, 0x00/255.0},
+            {0x4e/255.0, 0xaa/255.0, 0x90/255.0, 0xff/255.0},
+            {0x4e/255.0, 0xaa/255.0, 0x90/255.0, 0xff/255.0},
+            {0x00/255.0, 0x00/255.0, 0x00/255.0, 0x00/255.0}
+         },
+         {
+            {0x4e/255.0, 0xaa/255.0, 0x90/255.0, 0xff/255.0},
+            {0x29/255.0, 0xff/255.0, 0xff/255.0, 0xff/255.0},
+            {0x00/255.0, 0x00/255.0, 0x00/255.0, 0x00/255.0},
+            {0x4e/255.0, 0xaa/255.0, 0x90/255.0, 0xff/255.0}
+         },
+         {
+            {0x73/255.0, 0x55/255.0, 0x21/255.0, 0xff/255.0},
+            {0x00/255.0, 0x00/255.0, 0x00/255.0, 0x00/255.0},
+            {0x00/255.0, 0x00/255.0, 0x00/255.0, 0x00/255.0},
+            {0x4e/255.0, 0xaa/255.0, 0x90/255.0, 0xff/255.0}
+         },
+         {
+            {0x4e/255.0, 0xaa/255.0, 0x90/255.0, 0xff/255.0},
+            {0x00/255.0, 0x00/255.0, 0x00/255.0, 0x00/255.0},
+            {0x00/255.0, 0x00/255.0, 0x00/255.0, 0x00/255.0},
+            {0x4e/255.0, 0xaa/255.0, 0x90/255.0, 0xff/255.0}
+         }
+      }
+   },
+   {
+      PIPE_FORMAT_DXT3_RGBA,
+      {0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff},
+      {0xe7, 0x4a, 0x8f, 0x96, 0x5b, 0xc1, 0x1c, 0x84, 0xf6, 0x8f, 0xab, 0x32, 0x2a, 0x9a, 0x95, 0x5a},
+      {
+         {
+            {0x6d/255.0, 0xc6/255.0, 0x96/255.0, 0x77/255.0},
+            {0x6d/255.0, 0xc6/255.0, 0x96/255.0, 0xee/255.0},
+            {0x6d/255.0, 0xc6/255.0, 0x96/255.0, 0xaa/255.0},
+            {0x8c/255.0, 0xff/255.0, 0xb5/255.0, 0x44/255.0}
+         },
+         {
+            {0x6d/255.0, 0xc6/255.0, 0x96/255.0, 0xff/255.0},
+            {0x6d/255.0, 0xc6/255.0, 0x96/255.0, 0x88/255.0},
+            {0x31/255.0, 0x55/255.0, 0x5a/255.0, 0x66/255.0},
+            {0x6d/255.0, 0xc6/255.0, 0x96/255.0, 0x99/255.0}
+         },
+         {
+            {0x31/255.0, 0x55/255.0, 0x5a/255.0, 0xbb/255.0},
+            {0x31/255.0, 0x55/255.0, 0x5a/255.0, 0x55/255.0},
+            {0x31/255.0, 0x55/255.0, 0x5a/255.0, 0x11/255.0},
+            {0x6d/255.0, 0xc6/255.0, 0x96/255.0, 0xcc/255.0}
+         },
+         {
+            {0x6d/255.0, 0xc6/255.0, 0x96/255.0, 0xcc/255.0},
+            {0x6d/255.0, 0xc6/255.0, 0x96/255.0, 0x11/255.0},
+            {0x31/255.0, 0x55/255.0, 0x5a/255.0, 0x44/255.0},
+            {0x31/255.0, 0x55/255.0, 0x5a/255.0, 0x88/255.0}
+         }
+      }
+   },
+   {
+      PIPE_FORMAT_DXT5_RGBA,
+      {0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff},
+      {0xf8, 0x11, 0xc5, 0x0c, 0x9a, 0x73, 0xb4, 0x9c, 0xf6, 0x8f, 0xab, 0x32, 0x2a, 0x9a, 0x95, 0x5a},
+      {
+         {
+            {0x6d/255.0, 0xc6/255.0, 0x96/255.0, 0x74/255.0},
+            {0x6d/255.0, 0xc6/255.0, 0x96/255.0, 0xf8/255.0},
+            {0x6d/255.0, 0xc6/255.0, 0x96/255.0, 0xb6/255.0},
+            {0x8c/255.0, 0xff/255.0, 0xb5/255.0, 0x53/255.0}
+         },
+         {
+            {0x6d/255.0, 0xc6/255.0, 0x96/255.0, 0xf8/255.0},
+            {0x6d/255.0, 0xc6/255.0, 0x96/255.0, 0x95/255.0},
+            {0x31/255.0, 0x55/255.0, 0x5a/255.0, 0x53/255.0},
+            {0x6d/255.0, 0xc6/255.0, 0x96/255.0, 0x95/255.0}
+         },
+         {
+            {0x31/255.0, 0x55/255.0, 0x5a/255.0, 0xb6/255.0},
+            {0x31/255.0, 0x55/255.0, 0x5a/255.0, 0x53/255.0},
+            {0x31/255.0, 0x55/255.0, 0x5a/255.0, 0x11/255.0},
+            {0x6d/255.0, 0xc6/255.0, 0x96/255.0, 0xd7/255.0}
+         },
+         {
+            {0x6d/255.0, 0xc6/255.0, 0x96/255.0, 0xb6/255.0},
+            {0x6d/255.0, 0xc6/255.0, 0x96/255.0, 0x11/255.0},
+            {0x31/255.0, 0x55/255.0, 0x5a/255.0, 0x32/255.0},
+            {0x31/255.0, 0x55/255.0, 0x5a/255.0, 0x95/255.0}
+         }
+      }
+   },
+
 
    /*
     * Standard 8-bit integer formats