intel/blorp: make upload_shader() return a bool indicating success or failure
[mesa.git] / src / mesa / drivers / common / meta_blit.c
index d78300b802a8471a42c0147e69ebf78e1ae910da..991d52f21e9e1e12e73fe161447f5a94600aae33 100644 (file)
@@ -45,6 +45,7 @@
 #include "main/texenv.h"
 #include "main/teximage.h"
 #include "main/texparam.h"
+#include "main/uniforms.h"
 #include "main/varray.h"
 #include "main/viewport.h"
 #include "swrast/swrast.h"
@@ -69,24 +70,27 @@ setup_glsl_msaa_blit_scaled_shader(struct gl_context *ctx,
    const uint8_t *sample_map;
    char *sample_map_str = rzalloc_size(mem_ctx, 1);
    char *sample_map_expr = rzalloc_size(mem_ctx, 1);
-   char *texel_fetch_macro = rzalloc_size(mem_ctx, 1);;
-   const char *vs_source;
+   char *texel_fetch_macro = rzalloc_size(mem_ctx, 1);
    const char *sampler_array_suffix = "";
-   const char *texcoord_type = "vec2";
-   float y_scale;
+   float x_scale, y_scale;
    enum blit_msaa_shader shader_index;
 
    assert(src_rb);
    samples = MAX2(src_rb->NumSamples, 1);
-   y_scale = samples * 0.5;
+
+   if (samples == 16)
+      x_scale = 4.0;
+   else
+      x_scale = 2.0;
+   y_scale = samples / x_scale;
 
    /* We expect only power of 2 samples in source multisample buffer. */
-   assert((samples & (samples - 1)) == 0);
+   assert(samples > 0 && _mesa_is_pow_two(samples));
    while (samples >> (shader_offset + 1)) {
       shader_offset++;
    }
-   /* Update the assert if we plan to support more than 8X MSAA. */
-   assert(shader_offset > 0 && shader_offset < 4);
+   /* Update the assert if we plan to support more than 16X MSAA. */
+   assert(shader_offset > 0 && shader_offset <= 4);
 
    assert(target == GL_TEXTURE_2D_MULTISAMPLE ||
           target == GL_TEXTURE_2D_MULTISAMPLE_ARRAY);
@@ -98,18 +102,17 @@ setup_glsl_msaa_blit_scaled_shader(struct gl_context *ctx,
       shader_index += BLIT_2X_MSAA_SHADER_2D_MULTISAMPLE_ARRAY_SCALED_RESOLVE -
                       BLIT_2X_MSAA_SHADER_2D_MULTISAMPLE_SCALED_RESOLVE;
       sampler_array_suffix = "Array";
-      texcoord_type = "vec3";
    }
 
    if (blit->msaa_shaders[shader_index]) {
-      _mesa_UseProgram(blit->msaa_shaders[shader_index]);
+      _mesa_meta_use_program(ctx, blit->msaa_shaders[shader_index]);
       /* Update the uniform values. */
       loc_src_width =
-         glGetUniformLocation(blit->msaa_shaders[shader_index], "src_width");
+         _mesa_program_resource_location(blit->msaa_shaders[shader_index], GL_UNIFORM, "src_width");
       loc_src_height =
-         glGetUniformLocation(blit->msaa_shaders[shader_index], "src_height");
-      glUniform1f(loc_src_width, src_rb->Width);
-      glUniform1f(loc_src_height, src_rb->Height);
+         _mesa_program_resource_location(blit->msaa_shaders[shader_index], GL_UNIFORM, "src_height");
+      _mesa_Uniform1f(loc_src_width, src_rb->Width);
+      _mesa_Uniform1f(loc_src_height, src_rb->Height);
       return;
    }
 
@@ -131,8 +134,15 @@ setup_glsl_msaa_blit_scaled_shader(struct gl_context *ctx,
       sample_number =  "sample_map[int(2 * fract(coord.x) + 8 * fract(coord.y))]";
       sample_map = ctx->Const.SampleMap8x;
       break;
+   case 16:
+      sample_number =  "sample_map[int(4 * fract(coord.x) + 16 * fract(coord.y))]";
+      sample_map = ctx->Const.SampleMap16x;
+      break;
    default:
+      sample_number = NULL;
+      sample_map = NULL;
       _mesa_problem(ctx, "Unsupported sample count %d\n", samples);
+      unreachable("Unsupported sample count");
    }
 
    /* Create sample map string. */
@@ -146,45 +156,55 @@ setup_glsl_msaa_blit_scaled_shader(struct gl_context *ctx,
                           "   const int sample_map[%d] = int[%d](%s);\n",
                           samples, samples, sample_map_str);
 
-   ralloc_asprintf_append(&texel_fetch_macro,
-                          "#define TEXEL_FETCH(coord) texelFetch(texSampler, i%s(coord), %s);\n",
-                          texcoord_type, sample_number);
+   if (target == GL_TEXTURE_2D_MULTISAMPLE) {
+      ralloc_asprintf_append(&texel_fetch_macro,
+                             "#define TEXEL_FETCH(coord) texelFetch(texSampler, ivec2(coord), %s);\n",
+                             sample_number);
+   } else {
+      ralloc_asprintf_append(&texel_fetch_macro,
+                             "#define TEXEL_FETCH(coord) texelFetch(texSampler, ivec3(coord, layer), %s);\n",
+                             sample_number);
+   }
 
-   vs_source = ralloc_asprintf(mem_ctx,
+   static const char vs_source[] =
                                "#version 130\n"
-                               "in vec2 position;\n"
-                               "in %s textureCoords;\n"
-                               "out %s texCoords;\n"
+                               "#extension GL_ARB_explicit_attrib_location: enable\n"
+                               "layout(location = 0) in vec2 position;\n"
+                               "layout(location = 1) in vec3 textureCoords;\n"
+                               "out vec2 texCoords;\n"
+                               "flat out int layer;\n"
                                "void main()\n"
                                "{\n"
-                               "   texCoords = textureCoords;\n"
+                               "   texCoords = textureCoords.xy;\n"
+                               "   layer = int(textureCoords.z);\n"
                                "   gl_Position = vec4(position, 0.0, 1.0);\n"
-                               "}\n",
-                               texcoord_type,
-                               texcoord_type);
+                               "}\n"
+      ;
+
    fs_source = ralloc_asprintf(mem_ctx,
                                "#version 130\n"
                                "#extension GL_ARB_texture_multisample : enable\n"
                                "uniform sampler2DMS%s texSampler;\n"
                                "uniform float src_width, src_height;\n"
-                               "in %s texCoords;\n"
+                               "in vec2 texCoords;\n"
+                               "flat in int layer;\n"
                                "out vec4 out_color;\n"
                                "\n"
                                "void main()\n"
                                "{\n"
                                "%s"
                                "   vec2 interp;\n"
-                               "   const vec2 scale = vec2(2.0f, %ff);\n"
-                               "   const vec2 scale_inv = vec2(0.5f, %ff);\n"
-                               "   const vec2 s_0_offset = vec2(0.25f, %ff);\n"
+                               "   const vec2 scale = vec2(%ff, %ff);\n"
+                               "   const vec2 scale_inv = vec2(%ff, %ff);\n"
+                               "   const vec2 s_0_offset = vec2(%ff, %ff);\n"
                                "   vec2 s_0_coord, s_1_coord, s_2_coord, s_3_coord;\n"
                                "   vec4 s_0_color, s_1_color, s_2_color, s_3_color;\n"
                                "   vec4 x_0_color, x_1_color;\n"
                                "   vec2 tex_coord = texCoords - s_0_offset;\n"
                                "\n"
                                "   tex_coord *= scale;\n"
-                               "   clamp(tex_coord.x, 0.0f, scale.x * src_width - 1.0f);\n"
-                               "   clamp(tex_coord.y, 0.0f, scale.y * src_height - 1.0f);\n"
+                               "   tex_coord.x = clamp(tex_coord.x, 0.0f, scale.x * src_width - 1.0f);\n"
+                               "   tex_coord.y = clamp(tex_coord.y, 0.0f, scale.y * src_height - 1.0f);\n"
                                "   interp = fract(tex_coord);\n"
                                "   tex_coord = ivec2(tex_coord) * scale_inv;\n"
                                "\n"
@@ -203,26 +223,25 @@ setup_glsl_msaa_blit_scaled_shader(struct gl_context *ctx,
                                "#undef TEXEL_FETCH\n"
                                "\n"
                                "   /* Do bilinear filtering on sample colors. */\n"
-                               "   x_0_color =  mix(s_0_color, s_1_color, interp.x);\n"
-                               "   x_1_color =  mix(s_2_color, s_3_color, interp.x);\n"
+                               "   x_0_color = mix(s_0_color, s_1_color, interp.x);\n"
+                               "   x_1_color = mix(s_2_color, s_3_color, interp.x);\n"
                                "   out_color = mix(x_0_color, x_1_color, interp.y);\n"
                                "}\n",
                                sampler_array_suffix,
-                               texcoord_type,
                                sample_map_expr,
-                               y_scale,
-                               1.0f / y_scale,
-                               1.0f / samples,
+                               x_scale, y_scale,
+                               1.0f / x_scale, 1.0f / y_scale,
+                               0.5f / x_scale, 0.5f / y_scale,
                                texel_fetch_macro);
 
    _mesa_meta_compile_and_link_program(ctx, vs_source, fs_source, name,
                                        &blit->msaa_shaders[shader_index]);
    loc_src_width =
-      glGetUniformLocation(blit->msaa_shaders[shader_index], "src_width");
+      _mesa_program_resource_location(blit->msaa_shaders[shader_index], GL_UNIFORM, "src_width");
    loc_src_height =
-      glGetUniformLocation(blit->msaa_shaders[shader_index], "src_height");
-   glUniform1f(loc_src_width, src_rb->Width);
-   glUniform1f(loc_src_height, src_rb->Height);
+      _mesa_program_resource_location(blit->msaa_shaders[shader_index], GL_UNIFORM, "src_height");
+   _mesa_Uniform1f(loc_src_width, src_rb->Width);
+   _mesa_Uniform1f(loc_src_height, src_rb->Height);
 
    ralloc_free(mem_ctx);
 }
@@ -230,6 +249,7 @@ setup_glsl_msaa_blit_scaled_shader(struct gl_context *ctx,
 static void
 setup_glsl_msaa_blit_shader(struct gl_context *ctx,
                             struct blit_state *blit,
+                            const struct gl_framebuffer *drawFb,
                             struct gl_renderbuffer *src_rb,
                             GLenum target)
 {
@@ -258,14 +278,14 @@ setup_glsl_msaa_blit_shader(struct gl_context *ctx,
    }
 
    /* We expect only power of 2 samples in source multisample buffer. */
-   assert((samples & (samples - 1)) == 0);
+   assert(samples > 0 && _mesa_is_pow_two(samples));
    while (samples >> (shader_offset + 1)) {
       shader_offset++;
    }
    /* Update the assert if we plan to support more than 16X MSAA. */
    assert(shader_offset >= 0 && shader_offset <= 4);
 
-   if (ctx->DrawBuffer->Visual.samples > 1) {
+   if (drawFb->Visual.samples > 1) {
       /* If you're calling meta_BlitFramebuffer with the destination
        * multisampled, this is the only path that will work -- swrast and
        * CopyTexImage won't work on it either.
@@ -306,8 +326,8 @@ setup_glsl_msaa_blit_shader(struct gl_context *ctx,
       }
       break;
    default:
-      _mesa_problem(ctx, "Unkown texture target %s\n",
-                    _mesa_lookup_enum_by_nr(target));
+      _mesa_problem(ctx, "Unknown texture target %s\n",
+                    _mesa_enum_to_string(target));
       shader_index = BLIT_2X_MSAA_SHADER_2D_MULTISAMPLE_RESOLVE;
    }
 
@@ -327,7 +347,7 @@ setup_glsl_msaa_blit_shader(struct gl_context *ctx,
    }
 
    if (blit->msaa_shaders[shader_index]) {
-      _mesa_UseProgram(blit->msaa_shaders[shader_index]);
+      _mesa_meta_use_program(ctx, blit->msaa_shaders[shader_index]);
       return;
    }
 
@@ -338,17 +358,17 @@ setup_glsl_msaa_blit_shader(struct gl_context *ctx,
        shader_index == BLIT_MSAA_SHADER_2D_MULTISAMPLE_ARRAY_DEPTH_COPY ||
        shader_index == BLIT_MSAA_SHADER_2D_MULTISAMPLE_DEPTH_COPY) {
       char *sample_index;
-      const char *arb_sample_shading_extension_string;
+      const char *tex_coords = "texCoords";
 
       if (dst_is_msaa) {
-         arb_sample_shading_extension_string = "#extension GL_ARB_sample_shading : enable";
          sample_index = "gl_SampleID";
          name = "depth MSAA copy";
+
+         if (ctx->Extensions.ARB_gpu_shader5 && samples >= 16) {
+            /* See comment below for the color copy */
+            tex_coords = "interpolateAtOffset(texCoords, vec2(0.0))";
+         }
       } else {
-         /* Don't need that extension, since we're drawing to a single-sampled
-          * destination.
-          */
-         arb_sample_shading_extension_string = "";
          /* From the GL 4.3 spec:
           *
           *     "If there is a multisample buffer (the value of SAMPLE_BUFFERS
@@ -365,8 +385,9 @@ setup_glsl_msaa_blit_shader(struct gl_context *ctx,
 
       vs_source = ralloc_asprintf(mem_ctx,
                                   "#version 130\n"
-                                  "in vec2 position;\n"
-                                  "in %s textureCoords;\n"
+                                  "#extension GL_ARB_explicit_attrib_location: enable\n"
+                                  "layout(location = 0) in vec2 position;\n"
+                                  "layout(location = 1) in %s textureCoords;\n"
                                   "out %s texCoords;\n"
                                   "void main()\n"
                                   "{\n"
@@ -378,50 +399,82 @@ setup_glsl_msaa_blit_shader(struct gl_context *ctx,
       fs_source = ralloc_asprintf(mem_ctx,
                                   "#version 130\n"
                                   "#extension GL_ARB_texture_multisample : enable\n"
-                                  "%s\n"
+                                  "#extension GL_ARB_sample_shading : enable\n"
+                                  "#extension GL_ARB_gpu_shader5 : enable\n"
                                   "uniform sampler2DMS%s texSampler;\n"
                                   "in %s texCoords;\n"
                                   "out vec4 out_color;\n"
                                   "\n"
                                   "void main()\n"
                                   "{\n"
-                                  "   gl_FragDepth = texelFetch(texSampler, i%s(texCoords), %s).r;\n"
+                                  "   gl_FragDepth = texelFetch(texSampler, i%s(%s), %s).r;\n"
                                   "}\n",
-                                  arb_sample_shading_extension_string,
                                   sampler_array_suffix,
                                   texcoord_type,
                                   texcoord_type,
+                                  tex_coords,
                                   sample_index);
    } else {
       /* You can create 2D_MULTISAMPLE textures with 0 sample count (meaning 1
        * sample).  Yes, this is ridiculous.
        */
       char *sample_resolve;
-      const char *arb_sample_shading_extension_string;
       const char *merge_function;
       name = ralloc_asprintf(mem_ctx, "%svec4 MSAA %s",
                              vec4_prefix,
                              dst_is_msaa ? "copy" : "resolve");
 
       if (dst_is_msaa) {
-         arb_sample_shading_extension_string = "#extension GL_ARB_sample_shading : enable";
-         sample_resolve = ralloc_asprintf(mem_ctx, "   out_color = texelFetch(texSampler, i%s(texCoords), gl_SampleID);", texcoord_type);
+         const char *tex_coords;
+
+         if (ctx->Extensions.ARB_gpu_shader5 && samples >= 16) {
+            /* If interpolateAtOffset is available then it will be used to
+             * force the interpolation to the center. This is required at
+             * least on Intel hardware because it is possible to have a sample
+             * position on the 0 x or y axis which means it will lie exactly
+             * on the pixel boundary. If we let the hardware interpolate the
+             * coordinates at one of these positions then it is possible for
+             * it to jump to a neighboring texel when converting to ints due
+             * to rounding errors. This is only done for >= 16x MSAA because
+             * it probably has some overhead. It is more likely that some
+             * hardware will use one of these problematic positions at 16x
+             * MSAA because in that case in D3D they are defined to be at
+             * these positions.
+             */
+            tex_coords = "interpolateAtOffset(texCoords, vec2(0.0))";
+         } else {
+            tex_coords = "texCoords";
+         }
+
+         sample_resolve =
+            ralloc_asprintf(mem_ctx,
+                            "   out_color = texelFetch(texSampler, "
+                            "i%s(%s), gl_SampleID);",
+                            texcoord_type, tex_coords);
+
          merge_function = "";
       } else {
          int i;
          int step;
 
          if (src_datatype == GL_INT || src_datatype == GL_UNSIGNED_INT) {
-            merge_function =
-               "gvec4 merge(gvec4 a, gvec4 b) { return (a >> gvec4(1)) + (b >> gvec4(1)) + (a & b & gvec4(1)); }\n";
+            /* From the OpenGL ES 3.2 spec section 16.2.1:
+             *
+             *    "If the source formats are integer types or stencil values,
+             *    a single sample's value is selected for each pixel."
+             *
+             * The OpenGL 4.4 spec contains exactly the same language.
+             *
+             * We can accomplish this by making the merge function return just
+             * one of the two samples.  The compiler should do the rest.
+             */
+            merge_function = "gvec4 merge(gvec4 a, gvec4 b) { return a; }\n";
          } else {
             /* The divide will happen at the end for floats. */
             merge_function =
                "vec4 merge(vec4 a, vec4 b) { return (a + b); }\n";
          }
 
-         arb_sample_shading_extension_string = "";
-
          /* We're assuming power of two samples for this resolution procedure.
           *
           * To avoid losing any floating point precision if the samples all
@@ -429,7 +482,7 @@ setup_glsl_msaa_blit_shader(struct gl_context *ctx,
           * (so the floating point exponent just gets increased), rather than
           * doing a naive sum and dividing.
           */
-         assert((samples & (samples - 1)) == 0);
+         assert(_mesa_is_pow_two(samples));
          /* Fetch each individual sample. */
          sample_resolve = rzalloc_size(mem_ctx, 1);
          for (i = 0; i < samples; i++) {
@@ -464,8 +517,9 @@ setup_glsl_msaa_blit_shader(struct gl_context *ctx,
 
       vs_source = ralloc_asprintf(mem_ctx,
                                   "#version 130\n"
-                                  "in vec2 position;\n"
-                                  "in %s textureCoords;\n"
+                                  "#extension GL_ARB_explicit_attrib_location: enable\n"
+                                  "layout(location = 0) in vec2 position;\n"
+                                  "layout(location = 1) in %s textureCoords;\n"
                                   "out %s texCoords;\n"
                                   "void main()\n"
                                   "{\n"
@@ -477,7 +531,8 @@ setup_glsl_msaa_blit_shader(struct gl_context *ctx,
       fs_source = ralloc_asprintf(mem_ctx,
                                   "#version 130\n"
                                   "#extension GL_ARB_texture_multisample : enable\n"
-                                  "%s\n"
+                                  "#extension GL_ARB_sample_shading : enable\n"
+                                  "#extension GL_ARB_gpu_shader5 : enable\n"
                                   "#define gvec4 %svec4\n"
                                   "uniform %ssampler2DMS%s texSampler;\n"
                                   "in %s texCoords;\n"
@@ -488,7 +543,6 @@ setup_glsl_msaa_blit_shader(struct gl_context *ctx,
                                   "{\n"
                                   "%s\n" /* sample_resolve */
                                   "}\n",
-                                  arb_sample_shading_extension_string,
                                   vec4_prefix,
                                   vec4_prefix,
                                   sampler_array_suffix,
@@ -506,9 +560,11 @@ setup_glsl_msaa_blit_shader(struct gl_context *ctx,
 static void
 setup_glsl_blit_framebuffer(struct gl_context *ctx,
                             struct blit_state *blit,
+                            const struct gl_framebuffer *drawFb,
                             struct gl_renderbuffer *src_rb,
                             GLenum target, GLenum filter,
-                            bool is_scaled_blit)
+                            bool is_scaled_blit,
+                            bool do_depth)
 {
    unsigned texcoord_size;
    bool is_target_multisample = target == GL_TEXTURE_2D_MULTISAMPLE ||
@@ -521,15 +577,17 @@ setup_glsl_blit_framebuffer(struct gl_context *ctx,
 
    texcoord_size = 2 + (src_rb->Depth > 1 ? 1 : 0);
 
-   _mesa_meta_setup_vertex_objects(&blit->VAO, &blit->VBO, true,
+   _mesa_meta_setup_vertex_objects(ctx, &blit->VAO, &blit->buf_obj, true,
                                    2, texcoord_size, 0);
 
    if (is_target_multisample && is_filter_scaled_resolve && is_scaled_blit) {
       setup_glsl_msaa_blit_scaled_shader(ctx, blit, src_rb, target, filter);
    } else if (is_target_multisample) {
-      setup_glsl_msaa_blit_shader(ctx, blit, src_rb, target);
+      setup_glsl_msaa_blit_shader(ctx, blit, drawFb, src_rb, target);
    } else {
-      _mesa_meta_setup_blit_shader(ctx, target, &blit->shaders);
+      _mesa_meta_setup_blit_shader(ctx, target, do_depth,
+                                   do_depth ? &blit->shaders_with_depth
+                                            : &blit->shaders_without_depth);
    }
 }
 
@@ -541,12 +599,13 @@ setup_glsl_blit_framebuffer(struct gl_context *ctx,
  */
 static bool
 blitframebuffer_texture(struct gl_context *ctx,
+                        const struct gl_framebuffer *readFb,
+                        const struct gl_framebuffer *drawFb,
                         GLint srcX0, GLint srcY0, GLint srcX1, GLint srcY1,
                         GLint dstX0, GLint dstY0, GLint dstX1, GLint dstY1,
                         GLenum filter, GLint flipX, GLint flipY,
                         GLboolean glsl_version, GLboolean do_depth)
 {
-   const struct gl_framebuffer *readFb = ctx->ReadBuffer;
    int att_index = do_depth ? BUFFER_DEPTH : readFb->_ColorReadBufferIndex;
    const struct gl_renderbuffer_attachment *readAtt =
       &readFb->Attachment[att_index];
@@ -640,11 +699,13 @@ blitframebuffer_texture(struct gl_context *ctx,
    scaled_blit = dstW != srcW || dstH != srcH;
 
    if (glsl_version) {
-      setup_glsl_blit_framebuffer(ctx, blit, rb, target, filter, scaled_blit);
+      setup_glsl_blit_framebuffer(ctx, blit, drawFb, rb, target, filter, scaled_blit,
+                                  do_depth);
    }
    else {
-      _mesa_meta_setup_ff_tnl_for_blit(&ctx->Meta->Blit.VAO,
-                                       &ctx->Meta->Blit.VBO,
+      _mesa_meta_setup_ff_tnl_for_blit(ctx,
+                                       &ctx->Meta->Blit.VAO,
+                                       &ctx->Meta->Blit.buf_obj,
                                        2);
    }
 
@@ -654,37 +715,35 @@ blitframebuffer_texture(struct gl_context *ctx,
      printf("  srcTex %p  dstText %p\n", texObj, drawAtt->Texture);
    */
 
-   fb_tex_blit.sampler = _mesa_meta_setup_sampler(ctx, texObj, target, filter,
-                                                  srcLevel);
+   fb_tex_blit.samp_obj = _mesa_meta_setup_sampler(ctx, texObj, target, filter,
+                                                   srcLevel);
 
-   /* Always do our blits with no net sRGB decode or encode.
-    *
-    * However, if both the src and dst can be srgb decode/encoded, enable them
-    * so that we do any blending (from scaling or from MSAA resolves) in the
-    * right colorspace.
-    *
-    * Our choice of not doing any net encode/decode is from the GL 3.0
-    * specification:
-    *
-    *     "Blit operations bypass the fragment pipeline. The only fragment
-    *      operations which affect a blit are the pixel ownership test and the
-    *      scissor test."
-    *
-    * The GL 4.4 specification disagrees and says that the sRGB part of the
-    * fragment pipeline applies, but this was found to break applications.
-    */
    if (ctx->Extensions.EXT_texture_sRGB_decode) {
-      if (_mesa_get_format_color_encoding(rb->Format) == GL_SRGB &&
-          ctx->DrawBuffer->Visual.sRGBCapable) {
-         _mesa_SamplerParameteri(fb_tex_blit.sampler,
-                                 GL_TEXTURE_SRGB_DECODE_EXT, GL_DECODE_EXT);
-         _mesa_set_framebuffer_srgb(ctx, GL_TRUE);
-      } else {
-         _mesa_SamplerParameteri(fb_tex_blit.sampler,
-                                 GL_TEXTURE_SRGB_DECODE_EXT,
-                                 GL_SKIP_DECODE_EXT);
-         /* set_framebuffer_srgb was set by _mesa_meta_begin(). */
-      }
+      /* The GL 4.4 spec, section 18.3.1 ("Blitting Pixel Rectangles") says:
+       *
+       *    "When values are taken from the read buffer, if FRAMEBUFFER_SRGB
+       *     is enabled and the value of FRAMEBUFFER_ATTACHMENT_COLOR_ENCODING
+       *     for the framebuffer attachment corresponding to the read buffer
+       *     is SRGB (see section 9.2.3), the red, green, and blue components
+       *     are converted from the non-linear sRGB color space according to
+       *     equation 3.24.
+       *
+       *     When values are written to the draw buffers, blit operations
+       *     bypass most of the fragment pipeline.  The only fragment
+       *     operations which affect a blit are the pixel ownership test,
+       *     the scissor test, and sRGB conversion (see section 17.3.9)."
+       *
+       * ES 3.0 contains nearly the exact same text, but omits the part
+       * about GL_FRAMEBUFFER_SRGB as that doesn't exist in ES.  Mesa
+       * defaults it to on for ES contexts, so we can safely check it.
+       */
+      const bool decode =
+         ctx->Color.sRGBEnabled &&
+         _mesa_get_format_color_encoding(rb->Format) == GL_SRGB;
+
+      _mesa_set_sampler_srgb_decode(ctx, fb_tex_blit.samp_obj,
+                                    decode ? GL_DECODE_EXT
+                                           : GL_SKIP_DECODE_EXT);
    }
 
    if (!glsl_version) {
@@ -699,7 +758,7 @@ blitframebuffer_texture(struct gl_context *ctx,
 
       if (target == GL_TEXTURE_2D) {
          const struct gl_texture_image *texImage
-            = _mesa_select_tex_image(ctx, texObj, target, srcLevel);
+            = _mesa_select_tex_image(texObj, target, srcLevel);
          s0 = srcX0 / (float) texImage->Width;
          s1 = srcX1 / (float) texImage->Width;
          t0 = srcY0 / (float) texImage->Height;
@@ -741,7 +800,8 @@ blitframebuffer_texture(struct gl_context *ctx,
       verts[3].tex[1] = t1;
       verts[3].tex[2] = readAtt->Zoffset;
 
-      _mesa_BufferSubData(GL_ARRAY_BUFFER_ARB, 0, sizeof(verts), verts);
+      _mesa_buffer_sub_data(ctx, blit->buf_obj, 0, sizeof(verts), verts,
+                            __func__);
    }
 
    /* setup viewport */
@@ -758,12 +818,20 @@ blitframebuffer_texture(struct gl_context *ctx,
 }
 
 void
-_mesa_meta_fb_tex_blit_begin(const struct gl_context *ctx,
+_mesa_meta_fb_tex_blit_begin(struct gl_context *ctx,
                              struct fb_tex_blit_state *blit)
 {
-   blit->samplerSave =
-      ctx->Texture.Unit[ctx->Texture.CurrentUnit].Sampler ?
-      ctx->Texture.Unit[ctx->Texture.CurrentUnit].Sampler->Name : 0;
+   /* None of the existing callers preinitialize fb_tex_blit_state to zeros,
+    * and both use stack variables.  If samp_obj_save is not NULL,
+    * _mesa_reference_sampler_object will try to dereference it.  Leaving
+    * random garbage in samp_obj_save can only lead to crashes.
+    *
+    * Since the state isn't persistent across calls, we won't catch ref
+    * counting problems.
+    */
+   blit->samp_obj_save = NULL;
+   _mesa_reference_sampler_object(ctx, &blit->samp_obj_save,
+                                  ctx->Texture.Unit[ctx->Texture.CurrentUnit].Sampler);
    blit->tempTex = 0;
 }
 
@@ -771,26 +839,35 @@ void
 _mesa_meta_fb_tex_blit_end(struct gl_context *ctx, GLenum target,
                            struct fb_tex_blit_state *blit)
 {
+   struct gl_texture_object *const texObj =
+      _mesa_get_current_tex_object(ctx, target);
+
    /* Restore texture object state, the texture binding will
     * be restored by _mesa_meta_end().
     */
    if (target != GL_TEXTURE_RECTANGLE_ARB) {
-      _mesa_TexParameteri(target, GL_TEXTURE_BASE_LEVEL, blit->baseLevelSave);
-      _mesa_TexParameteri(target, GL_TEXTURE_MAX_LEVEL, blit->maxLevelSave);
+      _mesa_texture_parameteriv(ctx, texObj, GL_TEXTURE_BASE_LEVEL,
+                                &blit->baseLevelSave, false);
+      _mesa_texture_parameteriv(ctx, texObj, GL_TEXTURE_MAX_LEVEL,
+                                &blit->maxLevelSave, false);
+   }
 
-      if (ctx->Extensions.ARB_stencil_texturing) {
-         const struct gl_texture_object *texObj =
-            _mesa_get_current_tex_object(ctx, target);
+   /* If ARB_stencil_texturing is not supported, the mode won't have changed. */
+   if (texObj->StencilSampling != blit->stencilSamplingSave) {
+      /* GLint so the compiler won't complain about type signedness mismatch
+       * in the call to _mesa_texture_parameteriv below.
+       */
+      const GLint param = blit->stencilSamplingSave ?
+         GL_STENCIL_INDEX : GL_DEPTH_COMPONENT;
 
-         if (texObj->StencilSampling != blit->stencilSamplingSave)
-            _mesa_TexParameteri(target, GL_DEPTH_STENCIL_TEXTURE_MODE,
-                                blit->stencilSamplingSave ?
-                                   GL_STENCIL_INDEX : GL_DEPTH_COMPONENT);
-      }
+      _mesa_texture_parameteriv(ctx, texObj, GL_DEPTH_STENCIL_TEXTURE_MODE,
+                                &param, false);
    }
 
-   _mesa_BindSampler(ctx->Texture.CurrentUnit, blit->samplerSave);
-   _mesa_DeleteSamplers(1, &blit->sampler);
+   _mesa_bind_sampler(ctx, ctx->Texture.CurrentUnit, blit->samp_obj_save);
+   _mesa_reference_sampler_object(ctx, &blit->samp_obj_save, NULL);
+   _mesa_reference_sampler_object(ctx, &blit->samp_obj, NULL);
+
    if (blit->tempTex)
       _mesa_DeleteTextures(1, &blit->tempTex);
 }
@@ -834,31 +911,35 @@ _mesa_meta_bind_rb_as_tex_image(struct gl_context *ctx,
    return true;
 }
 
-GLuint
+struct gl_sampler_object *
 _mesa_meta_setup_sampler(struct gl_context *ctx,
-                         const struct gl_texture_object *texObj,
+                         struct gl_texture_object *texObj,
                          GLenum target, GLenum filter, GLuint srcLevel)
 {
-   GLuint sampler;
+   struct gl_sampler_object *samp_obj;
    GLenum tex_filter = (filter == GL_SCALED_RESOLVE_FASTEST_EXT ||
                         filter == GL_SCALED_RESOLVE_NICEST_EXT) ?
                        GL_NEAREST : filter;
 
-   _mesa_GenSamplers(1, &sampler);
-   _mesa_BindSampler(ctx->Texture.CurrentUnit, sampler);
+   samp_obj =  ctx->Driver.NewSamplerObject(ctx, 0xDEADBEEF);
+   if (samp_obj == NULL)
+      return NULL;
+
+   _mesa_bind_sampler(ctx, ctx->Texture.CurrentUnit, samp_obj);
+   _mesa_set_sampler_filters(ctx, samp_obj, tex_filter, tex_filter);
+   _mesa_set_sampler_wrap(ctx, samp_obj, GL_CLAMP_TO_EDGE, GL_CLAMP_TO_EDGE,
+                          samp_obj->WrapR);
 
    /* Prepare src texture state */
    _mesa_BindTexture(target, texObj->Name);
-   _mesa_SamplerParameteri(sampler, GL_TEXTURE_MIN_FILTER, tex_filter);
-   _mesa_SamplerParameteri(sampler, GL_TEXTURE_MAG_FILTER, tex_filter);
    if (target != GL_TEXTURE_RECTANGLE_ARB) {
-      _mesa_TexParameteri(target, GL_TEXTURE_BASE_LEVEL, srcLevel);
-      _mesa_TexParameteri(target, GL_TEXTURE_MAX_LEVEL, srcLevel);
+      _mesa_texture_parameteriv(ctx, texObj, GL_TEXTURE_BASE_LEVEL,
+                                (GLint *) &srcLevel, false);
+      _mesa_texture_parameteriv(ctx, texObj, GL_TEXTURE_MAX_LEVEL,
+                                (GLint *) &srcLevel, false);
    }
-   _mesa_SamplerParameteri(sampler, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
-   _mesa_SamplerParameteri(sampler, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
 
-   return sampler;
+   return samp_obj;
 }
 
 /**
@@ -867,6 +948,8 @@ _mesa_meta_setup_sampler(struct gl_context *ctx,
  */
 GLbitfield
 _mesa_meta_BlitFramebuffer(struct gl_context *ctx,
+                           const struct gl_framebuffer *readFb,
+                           const struct gl_framebuffer *drawFb,
                            GLint srcX0, GLint srcY0, GLint srcX1, GLint srcY1,
                            GLint dstX0, GLint dstY0, GLint dstX1, GLint dstY1,
                            GLbitfield mask, GLenum filter)
@@ -888,7 +971,7 @@ _mesa_meta_BlitFramebuffer(struct gl_context *ctx,
                                       ctx->Extensions.ARB_fragment_shader;
 
    /* Multisample texture blit support requires texture multisample. */
-   if (ctx->ReadBuffer->Visual.samples > 0 &&
+   if (readFb->Visual.samples > 0 &&
        !ctx->Extensions.ARB_texture_multisample) {
       return mask;
    }
@@ -896,16 +979,19 @@ _mesa_meta_BlitFramebuffer(struct gl_context *ctx,
    /* Clip a copy of the blit coordinates. If these differ from the input
     * coordinates, then we'll set the scissor.
     */
-   if (!_mesa_clip_blit(ctx, &clip.srcX0, &clip.srcY0, &clip.srcX1, &clip.srcY1,
+   if (!_mesa_clip_blit(ctx, readFb, drawFb,
+                        &clip.srcX0, &clip.srcY0, &clip.srcX1, &clip.srcY1,
                         &clip.dstX0, &clip.dstY0, &clip.dstX1, &clip.dstY1)) {
       /* clipped/scissored everything away */
       return 0;
    }
 
-   /* Only scissor affects blit, but we're doing to set a custom scissor if
-    * necessary anyway, so save/clear state.
+   /* Only scissor and FRAMEBUFFER_SRGB affect blit.  Leave sRGB alone, but
+    * save restore scissor as we'll set a custom scissor if necessary.
     */
-   _mesa_meta_begin(ctx, MESA_META_ALL & ~MESA_META_DRAW_BUFFERS);
+   _mesa_meta_begin(ctx, MESA_META_ALL &
+                         ~(MESA_META_DRAW_BUFFERS |
+                           MESA_META_FRAMEBUFFER_SRGB));
 
    /* Dithering shouldn't be performed for glBlitFramebuffer */
    _mesa_set_enable(ctx, GL_DITHER, GL_FALSE);
@@ -924,7 +1010,8 @@ _mesa_meta_BlitFramebuffer(struct gl_context *ctx,
 
    /* Try faster, direct texture approach first */
    if (mask & GL_COLOR_BUFFER_BIT) {
-      if (blitframebuffer_texture(ctx, srcX0, srcY0, srcX1, srcY1,
+      if (blitframebuffer_texture(ctx, readFb, drawFb,
+                                  srcX0, srcY0, srcX1, srcY1,
                                   dstX0, dstY0, dstX1, dstY1,
                                   filter, dstFlipX, dstFlipY,
                                   use_glsl_version, false)) {
@@ -933,7 +1020,8 @@ _mesa_meta_BlitFramebuffer(struct gl_context *ctx,
    }
 
    if (mask & GL_DEPTH_BUFFER_BIT && use_glsl_version) {
-      if (blitframebuffer_texture(ctx, srcX0, srcY0, srcX1, srcY1,
+      if (blitframebuffer_texture(ctx, readFb, drawFb,
+                                  srcX0, srcY0, srcX1, srcY1,
                                   dstX0, dstY0, dstX1, dstY1,
                                   filter, dstFlipX, dstFlipY,
                                   use_glsl_version, true)) {
@@ -951,16 +1039,16 @@ _mesa_meta_BlitFramebuffer(struct gl_context *ctx,
 }
 
 void
-_mesa_meta_glsl_blit_cleanup(struct blit_state *blit)
+_mesa_meta_glsl_blit_cleanup(struct gl_context *ctx, struct blit_state *blit)
 {
    if (blit->VAO) {
       _mesa_DeleteVertexArrays(1, &blit->VAO);
       blit->VAO = 0;
-      _mesa_DeleteBuffers(1, &blit->VBO);
-      blit->VBO = 0;
+      _mesa_reference_buffer_object(ctx, &blit->buf_obj, NULL);
    }
 
-   _mesa_meta_blit_shader_table_cleanup(&blit->shaders);
+   _mesa_meta_blit_shader_table_cleanup(ctx, &blit->shaders_with_depth);
+   _mesa_meta_blit_shader_table_cleanup(ctx, &blit->shaders_without_depth);
 
    _mesa_DeleteTextures(1, &blit->depthTex.TexObj);
    blit->depthTex.TexObj = 0;
@@ -968,20 +1056,22 @@ _mesa_meta_glsl_blit_cleanup(struct blit_state *blit)
 
 void
 _mesa_meta_and_swrast_BlitFramebuffer(struct gl_context *ctx,
+                                      struct gl_framebuffer *readFb,
+                                      struct gl_framebuffer *drawFb,
                                       GLint srcX0, GLint srcY0,
                                       GLint srcX1, GLint srcY1,
                                       GLint dstX0, GLint dstY0,
                                       GLint dstX1, GLint dstY1,
                                       GLbitfield mask, GLenum filter)
 {
-   mask = _mesa_meta_BlitFramebuffer(ctx,
+   mask = _mesa_meta_BlitFramebuffer(ctx, readFb, drawFb,
                                      srcX0, srcY0, srcX1, srcY1,
                                      dstX0, dstY0, dstX1, dstY1,
                                      mask, filter);
    if (mask == 0x0)
       return;
 
-   _swrast_BlitFramebuffer(ctx,
+   _swrast_BlitFramebuffer(ctx, readFb, drawFb,
                            srcX0, srcY0, srcX1, srcY1,
                            dstX0, dstY0, dstX1, dstY1,
                            mask, filter);