xserver mesa git update 25 Jul 2013

xserver          commit 43ac0491e36cdbb716b5c9d39c97d0aba3bebd75
mesa             commit 8a9df7a370b66ec50b6255e4d66ac1ed331319fb

5 files changed, 116 insertions, 51 deletions

diff --git a/mesalib/src/gallium/auxiliary/util/u_cpu_detect.c b/mesalib/src/gallium/auxiliary/util/u_cpu_detect.c
index b118fc8ae..588fc7c72 100644
--- a/mesalib/src/gallium/auxiliary/util/u_cpu_detect.c
+++ b/mesalib/src/gallium/auxiliary/util/u_cpu_detect.c
@@ -67,7 +67,7 @@
 
 #if defined(PIPE_OS_WINDOWS)
 #include <windows.h>
-#if defined(MSVC)
+#if defined(PIPE_CC_MSVC)
 #include <intrin.h>
 #endif
 #endif
@@ -211,6 +211,27 @@ cpuid(uint32_t ax, uint32_t *p)
    p[3] = 0;
 #endif
 }
+
+static INLINE uint64_t xgetbv(void)
+{
+#if defined(PIPE_CC_GCC)
+   uint32_t eax, edx;
+
+   __asm __volatile (
+     ".byte 0x0f, 0x01, 0xd0" // xgetbv isn't supported on gcc < 4.4
+     : "=a"(eax),
+       "=d"(edx)
+     : "c"(0)
+   );
+
+   return ((uint64_t)edx << 32) | eax;
+#elif defined(PIPE_CC_MSVC) && defined(_MSC_FULL_VER) && defined(_XCR_XFEATURE_ENABLED_MASK)
+   return _xgetbv(_XCR_XFEATURE_ENABLED_MASK);
+#else
+   return 0;
+#endif
+
+}
 #endif /* X86 or X86_64 */
 
 void
@@ -284,7 +305,9 @@ util_cpu_detect(void)
          util_cpu_caps.has_sse4_1 = (regs2[2] >> 19) & 1;
          util_cpu_caps.has_sse4_2 = (regs2[2] >> 20) & 1;
          util_cpu_caps.has_popcnt = (regs2[2] >> 23) & 1;
-         util_cpu_caps.has_avx    = (regs2[2] >> 28) & 1;
+         util_cpu_caps.has_avx    = ((regs2[2] >> 28) & 1) && // AVX
+                                    ((regs2[2] >> 27) & 1) && // OSXSAVE
+                                    ((xgetbv() & 6) == 6);    // XMM & YMM
          util_cpu_caps.has_f16c   = (regs2[2] >> 29) & 1;
          util_cpu_caps.has_mmx2   = util_cpu_caps.has_sse; /* SSE cpus supports mmxext too */
 
diff --git a/mesalib/src/mesa/drivers/common/meta.c b/mesalib/src/mesa/drivers/common/meta.c
index e8484299b..4a3497c9a 100644
--- a/mesalib/src/mesa/drivers/common/meta.c
+++ b/mesalib/src/mesa/drivers/common/meta.c
@@ -4095,10 +4095,29 @@ _mesa_meta_GetTexImage(struct gl_context *ctx,
        _mesa_get_format_datatype(texImage->TexFormat)
        == GL_UNSIGNED_NORMALIZED) {
       struct gl_texture_object *texObj = texImage->TexObject;
-      const GLuint slice = 0; /* only 2D compressed textures for now */
+      GLuint slice;
       /* Need to unlock the texture here to prevent deadlock... */
       _mesa_unlock_texture(ctx, texObj);
-      decompress_texture_image(ctx, texImage, slice, format, type, pixels);
+      for (slice = 0; slice < texImage->Depth; slice++) {
+         void *dst;
+         if (texImage->TexObject->Target == GL_TEXTURE_2D_ARRAY) {
+            /* Setup pixel packing.  SkipPixels and SkipRows will be applied
+             * in the decompress_texture_image() function's call to
+             * glReadPixels but we need to compute the dest slice's address
+             * here (according to SkipImages and ImageHeight).
+             */
+            struct gl_pixelstore_attrib packing = ctx->Pack;
+            packing.SkipPixels = 0;
+            packing.SkipRows = 0;
+            dst = _mesa_image_address3d(&packing, pixels, texImage->Width,
+                                        texImage->Height, format, type,
+                                        slice, 0, 0);
+         }
+         else {
+            dst = pixels;
+         }
+         decompress_texture_image(ctx, texImage, slice, format, type, dst);
+      }
       /* ... and relock it */
       _mesa_lock_texture(ctx, texObj);
    }
diff --git a/mesalib/src/mesa/main/mipmap.c b/mesalib/src/mesa/main/mipmap.c
index edf725721..583963232 100644
--- a/mesalib/src/mesa/main/mipmap.c
+++ b/mesalib/src/mesa/main/mipmap.c
@@ -2020,13 +2020,15 @@ generate_mipmap_compressed(struct gl_context *ctx, GLenum target,
    GLuint level;
    gl_format temp_format;
    GLint components;
-   GLuint temp_src_stride; /* in bytes */
+   GLuint temp_src_row_stride, temp_src_img_stride; /* in bytes */
    GLubyte *temp_src = NULL, *temp_dst = NULL;
    GLenum temp_datatype;
    GLenum temp_base_format;
+   GLubyte **temp_src_slices, **temp_dst_slices;
 
    /* only two types of compressed textures at this time */
    assert(texObj->Target == GL_TEXTURE_2D ||
+	  texObj->Target == GL_TEXTURE_2D_ARRAY ||
 	  texObj->Target == GL_TEXTURE_CUBE_MAP_ARB);
 
    /*
@@ -2051,15 +2053,24 @@ generate_mipmap_compressed(struct gl_context *ctx, GLenum target,
 
 
    /* allocate storage for the temporary, uncompressed image */
-   /* 20 extra bytes, just be safe when calling last FetchTexel */
-   temp_src_stride = _mesa_format_row_stride(temp_format, srcImage->Width);
-   temp_src = malloc(temp_src_stride * srcImage->Height + 20);
-   if (!temp_src) {
+   temp_src_row_stride = _mesa_format_row_stride(temp_format, srcImage->Width);
+   temp_src_img_stride = _mesa_format_image_size(temp_format, srcImage->Width,
+                                                 srcImage->Height, 1);
+   temp_src = malloc(temp_src_img_stride * srcImage->Depth);
+
+   /* Allocate storage for arrays of slice pointers */
+   temp_src_slices = malloc(srcImage->Depth * sizeof(GLubyte *));
+   temp_dst_slices = malloc(srcImage->Depth * sizeof(GLubyte *));
+
+   if (!temp_src || !temp_src_slices || !temp_dst_slices) {
+      free(temp_src);
+      free(temp_src_slices);
+      free(temp_dst_slices);
       _mesa_error(ctx, GL_OUT_OF_MEMORY, "generate mipmaps");
       return;
    }
 
-   /* decompress base image to the temporary */
+   /* decompress base image to the temporary src buffer */
    {
       /* save pixel packing mode */
       struct gl_pixelstore_attrib save = ctx->Pack;
@@ -2075,7 +2086,6 @@ generate_mipmap_compressed(struct gl_context *ctx, GLenum target,
       ctx->Pack = save;
    }
 
-
    for (level = texObj->BaseLevel; level < maxLevel; level++) {
       /* generate image[level+1] from image[level] */
       const struct gl_texture_image *srcImage;
@@ -2084,7 +2094,8 @@ generate_mipmap_compressed(struct gl_context *ctx, GLenum target,
       GLint dstWidth, dstHeight, dstDepth;
       GLint border;
       GLboolean nextLevel;
-      GLuint temp_dst_stride; /* in bytes */
+      GLuint temp_dst_row_stride, temp_dst_img_stride; /* in bytes */
+      GLuint i;
 
       /* get src image parameters */
       srcImage = _mesa_select_tex_image(ctx, texObj, target, level);
@@ -2100,9 +2111,12 @@ generate_mipmap_compressed(struct gl_context *ctx, GLenum target,
       if (!nextLevel)
 	 break;
 
-      temp_dst_stride = _mesa_format_row_stride(temp_format, dstWidth);
+      /* Compute dst image strides and alloc memory on first iteration */
+      temp_dst_row_stride = _mesa_format_row_stride(temp_format, dstWidth);
+      temp_dst_img_stride = _mesa_format_image_size(temp_format, dstWidth,
+                                                    dstHeight, 1);
       if (!temp_dst) {
-	 temp_dst = malloc(temp_dst_stride * dstHeight);
+	 temp_dst = malloc(temp_dst_img_stride * dstDepth);
 	 if (!temp_dst) {
 	    _mesa_error(ctx, GL_OUT_OF_MEMORY, "generate mipmaps");
 	    break;
@@ -2117,13 +2131,23 @@ generate_mipmap_compressed(struct gl_context *ctx, GLenum target,
          return;
       }
 
-      /* rescale src image to dest image */
+      /* for 2D arrays, setup array[depth] of slice pointers */
+      for (i = 0; i < srcDepth; i++) {
+         temp_src_slices[i] = temp_src + temp_src_img_stride * i;
+      }
+      for (i = 0; i < dstDepth; i++) {
+         temp_dst_slices[i] = temp_dst + temp_dst_img_stride * i;
+      }
+
+      /* Rescale src image to dest image.
+       * This will loop over the slices of a 2D array.
+       */
       _mesa_generate_mipmap_level(target, temp_datatype, components, border,
                                   srcWidth, srcHeight, srcDepth,
-                                  (const GLubyte **) &temp_src,
-                                  temp_src_stride,
+                                  (const GLubyte **) temp_src_slices,
+                                  temp_src_row_stride,
                                   dstWidth, dstHeight, dstDepth,
-                                  &temp_dst, temp_dst_stride);
+                                  temp_dst_slices, temp_dst_row_stride);
 
       if (!_mesa_prepare_mipmap_level(ctx, texObj, level + 1,
                                       dstWidth, dstHeight, dstDepth,
@@ -2135,7 +2159,7 @@ generate_mipmap_compressed(struct gl_context *ctx, GLenum target,
 
       /* The image space was allocated above so use glTexSubImage now */
       ctx->Driver.TexSubImage(ctx, 2, dstImage,
-                              0, 0, 0, dstWidth, dstHeight, 1,
+                              0, 0, 0, dstWidth, dstHeight, dstDepth,
                               temp_base_format, temp_datatype,
                               temp_dst, &ctx->DefaultPacking);
 
@@ -2144,12 +2168,15 @@ generate_mipmap_compressed(struct gl_context *ctx, GLenum target,
 	 GLubyte *temp = temp_src;
 	 temp_src = temp_dst;
 	 temp_dst = temp;
-	 temp_src_stride = temp_dst_stride;
+         temp_src_row_stride = temp_dst_row_stride;
+         temp_src_img_stride = temp_dst_img_stride;
       }
    } /* loop over mipmap levels */
 
    free(temp_src);
    free(temp_dst);
+   free(temp_src_slices);
+   free(temp_dst_slices);
 }
 
 /**
diff --git a/mesalib/src/mesa/main/texcompress_rgtc_tmp.h b/mesalib/src/mesa/main/texcompress_rgtc_tmp.h
index 277d69b17..5fa9de630 100644
--- a/mesalib/src/mesa/main/texcompress_rgtc_tmp.h
+++ b/mesalib/src/mesa/main/texcompress_rgtc_tmp.h
@@ -37,9 +37,9 @@ static void TAG(fetch_texel_rgtc)(unsigned srcRowStride, const TYPE *pixdata,
    const TYPE alpha0 = blksrc[0];
    const TYPE alpha1 = blksrc[1];
    const char bit_pos = ((j&3) * 4 + (i&3)) * 3;
-   const TYPE acodelow = blksrc[2 + bit_pos / 8];
-   const TYPE acodehigh = (3 + bit_pos / 8) < 8 ? blksrc[3 + bit_pos / 8] : 0;
-   const TYPE code = (acodelow >> (bit_pos & 0x7) |
+   const unsigned char acodelow = blksrc[2 + bit_pos / 8];
+   const unsigned char acodehigh = (3 + bit_pos / 8) < 8 ? blksrc[3 + bit_pos / 8] : 0;
+   const unsigned char code = (acodelow >> (bit_pos & 0x7) |
       (acodehigh  << (8 - (bit_pos & 0x7)))) & 0x7;
 
    if (code == 0)
diff --git a/mesalib/src/mesa/main/texgetimage.c b/mesalib/src/mesa/main/texgetimage.c
index 4c672e366..7050f1080 100644
--- a/mesalib/src/mesa/main/texgetimage.c
+++ b/mesalib/src/mesa/main/texgetimage.c
@@ -233,8 +233,8 @@ get_tex_rgba_compressed(struct gl_context *ctx, GLuint dimensions,
    const GLuint width = texImage->Width;
    const GLuint height = texImage->Height;
    const GLuint depth = texImage->Depth;
-   GLfloat *tempImage, *srcRow;
-   GLuint row;
+   GLfloat *tempImage, *tempSlice, *srcRow;
+   GLuint row, slice;
 
    /* Decompress into temp float buffer, then pack into user buffer */
    tempImage = malloc(width * height * depth
@@ -244,20 +244,22 @@ get_tex_rgba_compressed(struct gl_context *ctx, GLuint dimensions,
       return;
    }
 
-   /* Decompress the texture image - results in 'tempImage' */
-   {
+   /* Decompress the texture image slices - results in 'tempImage' */
+   for (slice = 0; slice < depth; slice++) {
       GLubyte *srcMap;
       GLint srcRowStride;
 
-      ctx->Driver.MapTextureImage(ctx, texImage, 0,
+      tempSlice = tempImage + slice * 4 * width * height;
+
+      ctx->Driver.MapTextureImage(ctx, texImage, slice,
                                   0, 0, width, height,
                                   GL_MAP_READ_BIT,
                                   &srcMap, &srcRowStride);
       if (srcMap) {
          _mesa_decompress_image(texFormat, width, height,
-                                srcMap, srcRowStride, tempImage);
+                                srcMap, srcRowStride, tempSlice);
 
-         ctx->Driver.UnmapTextureImage(ctx, texImage, 0);
+         ctx->Driver.UnmapTextureImage(ctx, texImage, slice);
       }
       else {
          _mesa_error(ctx, GL_OUT_OF_MEMORY, "glGetTexImage");
@@ -294,15 +296,19 @@ get_tex_rgba_compressed(struct gl_context *ctx, GLuint dimensions,
                               rebaseFormat);
    }
 
-   srcRow = tempImage;
-   for (row = 0; row < height; row++) {
-      void *dest = _mesa_image_address(dimensions, &ctx->Pack, pixels,
-                                       width, height, format, type,
-                                       0, row, 0);
-
-      _mesa_pack_rgba_span_float(ctx, width, (GLfloat (*)[4]) srcRow,
-                                 format, type, dest, &ctx->Pack, transferOps);
-      srcRow += width * 4;
+   tempSlice = tempImage;
+   for (slice = 0; slice < depth; slice++) {
+      srcRow = tempSlice;
+      for (row = 0; row < height; row++) {
+         void *dest = _mesa_image_address(dimensions, &ctx->Pack, pixels,
+                                          width, height, format, type,
+                                          slice, row, 0);
+
+         _mesa_pack_rgba_span_float(ctx, width, (GLfloat (*)[4]) srcRow,
+                                    format, type, dest, &ctx->Pack, transferOps);
+         srcRow += 4 * width;
+      }
+      tempSlice += 4 * width * height;
    }
 
    free(tempImage);
@@ -616,18 +622,8 @@ _mesa_get_teximage(struct gl_context *ctx,
                    GLenum format, GLenum type, GLvoid *pixels,
                    struct gl_texture_image *texImage)
 {
-   GLuint dimensions;
-
-   switch (texImage->TexObject->Target) {
-   case GL_TEXTURE_1D:
-      dimensions = 1;
-      break;
-   case GL_TEXTURE_3D:
-      dimensions = 3;
-      break;
-   default:
-      dimensions = 2;
-   }
+   const GLuint dimensions =
+      _mesa_get_texture_dimensions(texImage->TexObject->Target);
 
    /* map dest buffer, if PBO */
    if (_mesa_is_bufferobj(ctx->Pack.BufferObj)) {