mesa/st: Reuse st_choose_matching_format from st_choose_format().
[mesa.git] / src / mesa / state_tracker / st_cb_drawpixels.c
index 04a9de06520dd89364e243e1991f8a3a4907fe7b..6b47548d33ba7ab7355a3bf4c5a12728d7fe9bef 100644 (file)
   *   Brian Paul
   */
 
+#include "main/errors.h"
 #include "main/imports.h"
 #include "main/image.h"
 #include "main/bufferobj.h"
 #include "main/blit.h"
 #include "main/format_pack.h"
+#include "main/framebuffer.h"
 #include "main/macros.h"
 #include "main/mtypes.h"
 #include "main/pack.h"
 #include "main/pbo.h"
 #include "main/readpix.h"
+#include "main/state.h"
 #include "main/texformat.h"
 #include "main/teximage.h"
 #include "main/texstore.h"
 #include "st_cb_fbo.h"
 #include "st_context.h"
 #include "st_debug.h"
+#include "st_draw.h"
 #include "st_format.h"
 #include "st_program.h"
+#include "st_sampler_view.h"
+#include "st_scissor.h"
 #include "st_texture.h"
+#include "st_util.h"
+#include "st_nir.h"
 
 #include "pipe/p_context.h"
 #include "pipe/p_defines.h"
 #include "tgsi/tgsi_ureg.h"
-#include "util/u_draw_quad.h"
-#include "util/u_format.h"
+#include "util/format/u_format.h"
 #include "util/u_inlines.h"
 #include "util/u_math.h"
+#include "util/u_simple_shaders.h"
 #include "util/u_tile.h"
-#include "util/u_upload_mgr.h"
 #include "cso_cache/cso_context.h"
 
+#include "compiler/nir/nir_builder.h"
 
 /**
- * Create fragment program that does a TEX() instruction to get a Z and/or
- * stencil value value, then writes to FRAG_RESULT_DEPTH/FRAG_RESULT_STENCIL.
- * Used for glDrawPixels(GL_DEPTH_COMPONENT / GL_STENCIL_INDEX).
- * Pass fragment color through as-is.
+ * We have a simple glDrawPixels cache to try to optimize the case where the
+ * same image is drawn over and over again.  It basically works as follows:
  *
- * \return CSO of the fragment shader.
+ * 1. After we construct a texture map with the image and draw it, we do
+ *    not discard the texture.  We keep it around, plus we note the
+ *    glDrawPixels width, height, format, etc. parameters and keep a copy
+ *    of the image in a malloc'd buffer.
+ *
+ * 2. On the next glDrawPixels we check if the parameters match the previous
+ *    call.  If those match, we check if the image matches the previous image
+ *    via a memcmp() call.  If everything matches, we re-use the previous
+ *    texture, thereby avoiding the cost creating a new texture and copying
+ *    the image to it.
+ *
+ * The effectiveness of this cache depends upon:
+ * 1. If the memcmp() finds a difference, it happens relatively quickly.
+      Hopefully, not just the last pixels differ!
+ * 2. If the memcmp() finds no difference, doing that check is faster than
+ *    creating and loading a texture.
+ *
+ * Notes:
+ * 1. We don't support any pixel unpacking parameters.
+ * 2. We don't try to cache images in Pixel Buffer Objects.
+ * 3. Instead of saving the whole image, perhaps some sort of reliable
+ *    checksum function could be used instead.
  */
+#define USE_DRAWPIXELS_CACHE 1
+
+static nir_ssa_def *
+sample_via_nir(nir_builder *b, nir_variable *texcoord,
+               const char *name, int sampler)
+{
+   const struct glsl_type *sampler2D =
+      glsl_sampler_type(GLSL_SAMPLER_DIM_2D, false, false, GLSL_TYPE_FLOAT);
+
+   nir_variable *var =
+      nir_variable_create(b->shader, nir_var_uniform, sampler2D, name);
+   var->data.binding = sampler;
+   var->data.explicit_binding = true;
+
+   nir_deref_instr *deref = nir_build_deref_var(b, var);
+
+   nir_tex_instr *tex = nir_tex_instr_create(b->shader, 3);
+   tex->op = nir_texop_tex;
+   tex->sampler_dim = GLSL_SAMPLER_DIM_2D;
+   tex->coord_components = 2;
+   tex->dest_type = nir_type_float;
+   tex->src[0].src_type = nir_tex_src_texture_deref;
+   tex->src[0].src = nir_src_for_ssa(&deref->dest.ssa);
+   tex->src[1].src_type = nir_tex_src_sampler_deref;
+   tex->src[1].src = nir_src_for_ssa(&deref->dest.ssa);
+   tex->src[2].src_type = nir_tex_src_coord;
+   tex->src[2].src =
+      nir_src_for_ssa(nir_channels(b, nir_load_var(b, texcoord),
+                                   (1 << tex->coord_components) - 1));
+
+   nir_ssa_dest_init(&tex->instr, &tex->dest, 4, 32, NULL);
+   nir_builder_instr_insert(b, &tex->instr);
+   return nir_channel(b, &tex->dest.ssa, 0);
+}
+
 static void *
-get_drawpix_z_stencil_program(struct st_context *st,
-                              GLboolean write_depth,
-                              GLboolean write_stencil)
+make_drawpix_z_stencil_program_nir(struct st_context *st,
+                                   bool write_depth,
+                                   bool write_stencil)
+{
+   struct nir_builder b;
+   const nir_shader_compiler_options *options =
+      st->ctx->Const.ShaderCompilerOptions[MESA_SHADER_FRAGMENT].NirOptions;
+
+   nir_builder_init_simple_shader(&b, NULL, MESA_SHADER_FRAGMENT, options);
+
+   nir_variable *texcoord =
+      nir_variable_create(b.shader, nir_var_shader_in, glsl_vec_type(2),
+                          "texcoord");
+   texcoord->data.location = VARYING_SLOT_TEX0;
+
+   if (write_depth) {
+      nir_variable *out =
+         nir_variable_create(b.shader, nir_var_shader_out, glsl_float_type(),
+                             "gl_FragDepth");
+      out->data.location = FRAG_RESULT_DEPTH;
+      nir_ssa_def *depth = sample_via_nir(&b, texcoord, "depth", 0);
+      nir_store_var(&b, out, depth, 0x1);
+
+      /* Also copy color */
+      nir_variable *color_in =
+         nir_variable_create(b.shader, nir_var_shader_in, glsl_vec_type(4),
+                             "v_color");
+      color_in->data.location = VARYING_SLOT_COL0;
+
+      nir_variable *color_out =
+         nir_variable_create(b.shader, nir_var_shader_out, glsl_vec_type(4),
+                             "gl_FragColor");
+      color_out->data.location = FRAG_RESULT_COLOR;
+      nir_copy_var(&b, color_out, color_in);
+   }
+
+   if (write_stencil) {
+      nir_variable *out =
+         nir_variable_create(b.shader, nir_var_shader_out, glsl_uint_type(),
+                             "gl_FragStencilRefARB");
+      out->data.location = FRAG_RESULT_STENCIL;
+      nir_ssa_def *stencil = sample_via_nir(&b, texcoord, "stencil", 1);
+      nir_store_var(&b, out, stencil, 0x1);
+   }
+
+   char name[14];
+   snprintf(name, 14, "drawpixels %s%s",
+            write_depth ? "Z" : "", write_stencil ? "S" : "");
+
+   return st_nir_finish_builtin_shader(st, b.shader, name);
+}
+
+
+static void *
+make_drawpix_z_stencil_program_tgsi(struct st_context *st,
+                                    bool write_depth,
+                                    bool write_stencil)
 {
    struct ureg_program *ureg;
    struct ureg_src depth_sampler, stencil_sampler;
    struct ureg_src texcoord, color;
    struct ureg_dst out_color, out_depth, out_stencil;
-   const GLuint shaderIndex = write_depth * 2 + write_stencil;
-   void *cso;
-
-   assert(shaderIndex < ARRAY_SIZE(st->drawpix.zs_shaders));
-
-   if (st->drawpix.zs_shaders[shaderIndex]) {
-      /* already have the proper shader */
-      return st->drawpix.zs_shaders[shaderIndex];
-   }
 
-   ureg = ureg_create(TGSI_PROCESSOR_FRAGMENT);
+   ureg = ureg_create(PIPE_SHADER_FRAGMENT);
    if (ureg == NULL)
       return NULL;
 
@@ -111,11 +218,21 @@ get_drawpix_z_stencil_program(struct st_context *st,
       out_color = ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0);
 
       depth_sampler = ureg_DECL_sampler(ureg, 0);
+      ureg_DECL_sampler_view(ureg, 0, TGSI_TEXTURE_2D,
+                             TGSI_RETURN_TYPE_FLOAT,
+                             TGSI_RETURN_TYPE_FLOAT,
+                             TGSI_RETURN_TYPE_FLOAT,
+                             TGSI_RETURN_TYPE_FLOAT);
       out_depth = ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 0);
    }
 
    if (write_stencil) {
       stencil_sampler = ureg_DECL_sampler(ureg, 1);
+      ureg_DECL_sampler_view(ureg, 1, TGSI_TEXTURE_2D,
+                             TGSI_RETURN_TYPE_UINT,
+                             TGSI_RETURN_TYPE_UINT,
+                             TGSI_RETURN_TYPE_UINT,
+                             TGSI_RETURN_TYPE_UINT);
       out_stencil = ureg_DECL_output(ureg, TGSI_SEMANTIC_STENCIL, 0);
    }
 
@@ -136,7 +253,42 @@ get_drawpix_z_stencil_program(struct st_context *st,
                TGSI_TEXTURE_2D, texcoord, stencil_sampler);
 
    ureg_END(ureg);
-   cso = ureg_create_shader_and_destroy(ureg, st->pipe);
+   return ureg_create_shader_and_destroy(ureg, st->pipe);
+}
+
+
+/**
+ * Create fragment program that does a TEX() instruction to get a Z and/or
+ * stencil value value, then writes to FRAG_RESULT_DEPTH/FRAG_RESULT_STENCIL.
+ * Used for glDrawPixels(GL_DEPTH_COMPONENT / GL_STENCIL_INDEX).
+ * Pass fragment color through as-is.
+ *
+ * \return CSO of the fragment shader.
+ */
+static void *
+get_drawpix_z_stencil_program(struct st_context *st,
+                              bool write_depth,
+                              bool write_stencil)
+{
+   struct pipe_screen *pscreen = st->pipe->screen;
+   const GLuint shaderIndex = write_depth * 2 + write_stencil;
+   void *cso;
+
+   assert(shaderIndex < ARRAY_SIZE(st->drawpix.zs_shaders));
+
+   if (st->drawpix.zs_shaders[shaderIndex]) {
+      /* already have the proper shader */
+      return st->drawpix.zs_shaders[shaderIndex];
+   }
+
+   enum pipe_shader_ir preferred_ir =
+      pscreen->get_shader_param(pscreen, PIPE_SHADER_FRAGMENT,
+                                PIPE_SHADER_CAP_PREFERRED_IR);
+
+   if (preferred_ir == PIPE_SHADER_IR_NIR)
+      cso = make_drawpix_z_stencil_program_nir(st, write_depth, write_stencil);
+   else
+      cso = make_drawpix_z_stencil_program_tgsi(st, write_depth, write_stencil);
 
    /* save the new shader */
    st->drawpix.zs_shaders[shaderIndex] = cso;
@@ -146,45 +298,44 @@ get_drawpix_z_stencil_program(struct st_context *st,
 
 /**
  * Create a simple vertex shader that just passes through the
- * vertex position and texcoord (and optionally, color).
+ * vertex position, texcoord, and color.
  */
-static void *
-make_passthrough_vertex_shader(struct st_context *st, 
-                               GLboolean passColor)
+void
+st_make_passthrough_vertex_shader(struct st_context *st)
 {
-   const unsigned texcoord_semantic = st->needs_texcoord_semantic ?
-      TGSI_SEMANTIC_TEXCOORD : TGSI_SEMANTIC_GENERIC;
-
-   if (!st->drawpix.vert_shaders[passColor]) {
-      struct ureg_program *ureg = ureg_create( TGSI_PROCESSOR_VERTEX );
-
-      if (ureg == NULL)
-         return NULL;
+   struct pipe_context *pipe = st->pipe;
+   struct pipe_screen *screen = pipe->screen;
 
-      /* MOV result.pos, vertex.pos; */
-      ureg_MOV(ureg, 
-               ureg_DECL_output( ureg, TGSI_SEMANTIC_POSITION, 0 ),
-               ureg_DECL_vs_input( ureg, 0 ));
-      
-      /* MOV result.texcoord0, vertex.attr[1]; */
-      ureg_MOV(ureg, 
-               ureg_DECL_output( ureg, texcoord_semantic, 0 ),
-               ureg_DECL_vs_input( ureg, 1 ));
-      
-      if (passColor) {
-         /* MOV result.color0, vertex.attr[2]; */
-         ureg_MOV(ureg, 
-                  ureg_DECL_output( ureg, TGSI_SEMANTIC_COLOR, 0 ),
-                  ureg_DECL_vs_input( ureg, 2 ));
-      }
+   if (st->passthrough_vs)
+      return;
 
-      ureg_END( ureg );
-      
-      st->drawpix.vert_shaders[passColor] = 
-         ureg_create_shader_and_destroy( ureg, st->pipe );
+   enum pipe_shader_ir preferred_ir =
+      screen->get_shader_param(screen, PIPE_SHADER_VERTEX,
+                               PIPE_SHADER_CAP_PREFERRED_IR);
+
+   if (preferred_ir == PIPE_SHADER_IR_NIR) {
+      unsigned inputs[] =
+         {  VERT_ATTRIB_POS, VERT_ATTRIB_COLOR0, VERT_ATTRIB_GENERIC0 };
+      unsigned outputs[] =
+         { VARYING_SLOT_POS,  VARYING_SLOT_COL0,    VARYING_SLOT_TEX0 };
+
+      st->passthrough_vs =
+         st_nir_make_passthrough_shader(st, "drawpixels VS",
+                                        MESA_SHADER_VERTEX, 3,
+                                        inputs, outputs, NULL, 0);
+   } else {
+      const enum tgsi_semantic semantic_names[] = {
+         TGSI_SEMANTIC_POSITION,
+         TGSI_SEMANTIC_COLOR,
+         st->needs_texcoord_semantic ? TGSI_SEMANTIC_TEXCOORD :
+                                       TGSI_SEMANTIC_GENERIC
+      };
+      const uint semantic_indexes[] = { 0, 0, 0 };
+
+      st->passthrough_vs =
+         util_make_vertex_passthrough_shader(st->pipe, 3, semantic_names,
+                                             semantic_indexes, false);
    }
-
-   return st->drawpix.vert_shaders[passColor];
 }
 
 
@@ -330,6 +481,130 @@ alloc_texture(struct st_context *st, GLsizei width, GLsizei height,
 }
 
 
+/**
+ * Search the cache for an image which matches the given parameters.
+ * \return  pipe_resource pointer if found, NULL if not found.
+ */
+static struct pipe_resource *
+search_drawpixels_cache(struct st_context *st,
+                        GLsizei width, GLsizei height,
+                        GLenum format, GLenum type,
+                        const struct gl_pixelstore_attrib *unpack,
+                        const void *pixels)
+{
+   struct pipe_resource *pt = NULL;
+   const GLint bpp = _mesa_bytes_per_pixel(format, type);
+   unsigned i;
+
+   if ((unpack->RowLength != 0 && unpack->RowLength != width) ||
+       unpack->SkipPixels != 0 ||
+       unpack->SkipRows != 0 ||
+       unpack->SwapBytes ||
+       _mesa_is_bufferobj(unpack->BufferObj)) {
+      /* we don't allow non-default pixel unpacking values */
+      return NULL;
+   }
+
+   /* Search cache entries for a match */
+   for (i = 0; i < ARRAY_SIZE(st->drawpix_cache.entries); i++) {
+      struct drawpix_cache_entry *entry = &st->drawpix_cache.entries[i];
+
+      if (width == entry->width &&
+          height == entry->height &&
+          format == entry->format &&
+          type == entry->type &&
+          pixels == entry->user_pointer &&
+          entry->image) {
+         assert(entry->texture);
+
+         /* check if the pixel data is the same */
+         if (memcmp(pixels, entry->image, width * height * bpp) == 0) {
+            /* Success - found a cache match */
+            pipe_resource_reference(&pt, entry->texture);
+            /* refcount of returned texture should be at least two here.  One
+             * reference for the cache to hold on to, one for the caller (which
+             * it will release), and possibly more held by the driver.
+             */
+            assert(pt->reference.count >= 2);
+
+            /* update the age of this entry */
+            entry->age = ++st->drawpix_cache.age;
+
+            return pt;
+         }
+      }
+   }
+
+   /* no cache match found */
+   return NULL;
+}
+
+
+/**
+ * Find the oldest entry in the glDrawPixels cache.  We'll replace this
+ * one when we need to store a new image.
+ */
+static struct drawpix_cache_entry *
+find_oldest_drawpixels_cache_entry(struct st_context *st)
+{
+   unsigned oldest_age = ~0u, oldest_index = ~0u;
+   unsigned i;
+
+   /* Find entry with oldest (lowest) age */
+   for (i = 0; i < ARRAY_SIZE(st->drawpix_cache.entries); i++) {
+      const struct drawpix_cache_entry *entry = &st->drawpix_cache.entries[i];
+      if (entry->age < oldest_age) {
+         oldest_age = entry->age;
+         oldest_index = i;
+      }
+   }
+
+   assert(oldest_index != ~0u);
+
+   return &st->drawpix_cache.entries[oldest_index];
+}
+
+
+/**
+ * Try to save the given glDrawPixels image in the cache.
+ */
+static void
+cache_drawpixels_image(struct st_context *st,
+                       GLsizei width, GLsizei height,
+                       GLenum format, GLenum type,
+                       const struct gl_pixelstore_attrib *unpack,
+                       const void *pixels,
+                       struct pipe_resource *pt)
+{
+   if ((unpack->RowLength == 0 || unpack->RowLength == width) &&
+       unpack->SkipPixels == 0 &&
+       unpack->SkipRows == 0) {
+      const GLint bpp = _mesa_bytes_per_pixel(format, type);
+      struct drawpix_cache_entry *entry =
+         find_oldest_drawpixels_cache_entry(st);
+      assert(entry);
+      entry->width = width;
+      entry->height = height;
+      entry->format = format;
+      entry->type = type;
+      entry->user_pointer = pixels;
+      free(entry->image);
+      entry->image = malloc(width * height * bpp);
+      if (entry->image) {
+         memcpy(entry->image, pixels, width * height * bpp);
+         pipe_resource_reference(&entry->texture, pt);
+         entry->age = ++st->drawpix_cache.age;
+      }
+      else {
+         /* out of memory, free/disable cached texture */
+         entry->width = 0;
+         entry->height = 0;
+         pipe_resource_reference(&entry->texture, NULL);
+      }
+   }
+}
+
+
 /**
  * Make texture containing an image for glDrawPixels image.
  * If 'pixels' is NULL, leave the texture image data undefined.
@@ -338,15 +613,23 @@ static struct pipe_resource *
 make_texture(struct st_context *st,
             GLsizei width, GLsizei height, GLenum format, GLenum type,
             const struct gl_pixelstore_attrib *unpack,
-            const GLvoid *pixels)
+            const void *pixels)
 {
    struct gl_context *ctx = st->ctx;
    struct pipe_context *pipe = st->pipe;
    mesa_format mformat;
-   struct pipe_resource *pt;
+   struct pipe_resource *pt = NULL;
    enum pipe_format pipeFormat;
    GLenum baseInternalFormat;
 
+#if USE_DRAWPIXELS_CACHE
+   pt = search_drawpixels_cache(st, width, height, format, type,
+                                unpack, pixels);
+   if (pt) {
+      return pt;
+   }
+#endif
+
    /* Choose a pixel format for the temp texture which will hold the
     * image to draw.
     */
@@ -358,8 +641,9 @@ make_texture(struct st_context *st,
       GLenum intFormat = internal_format(ctx, format, type);
 
       pipeFormat = st_choose_format(st, intFormat, format, type,
-                                    PIPE_TEXTURE_2D, 0, PIPE_BIND_SAMPLER_VIEW,
-                                    FALSE);
+                                    st->internal_target, 0, 0,
+                                    PIPE_BIND_SAMPLER_VIEW,
+                                    false, false);
       assert(pipeFormat != PIPE_FORMAT_NONE);
    }
 
@@ -379,7 +663,6 @@ make_texture(struct st_context *st,
 
    {
       struct pipe_transfer *transfer;
-      GLboolean success;
       GLubyte *dest;
       const GLbitfield imageTransferStateSave = ctx->_ImageTransferState;
 
@@ -390,7 +673,11 @@ make_texture(struct st_context *st,
       dest = pipe_transfer_map(pipe, pt, 0, 0,
                                PIPE_TRANSFER_WRITE, 0, 0,
                                width, height, &transfer);
-
+      if (!dest) {
+         pipe_resource_reference(&pt, NULL);
+         _mesa_unmap_pbo_source(ctx, unpack);
+         return NULL;
+      }
 
       /* Put image into texture transfer.
        * Note that the image is actually going to be upside down in
@@ -412,9 +699,9 @@ make_texture(struct st_context *st,
                               format, type,     /* src format/type */
                               pixels,           /* data source */
                               unpack);
-         success = GL_TRUE;
       }
       else {
+         ASSERTED bool success;
          success = _mesa_texstore(ctx, 2,           /* dims */
                                   baseInternalFormat, /* baseInternalFormat */
                                   mformat,          /* mesa_format */
@@ -424,120 +711,27 @@ make_texture(struct st_context *st,
                                   format, type,     /* src format/type */
                                   pixels,           /* data source */
                                   unpack);
+
+         assert(success);
       }
 
       /* unmap */
       pipe_transfer_unmap(pipe, transfer);
 
-      assert(success);
-
       /* restore */
       ctx->_ImageTransferState = imageTransferStateSave;
    }
 
    _mesa_unmap_pbo_source(ctx, unpack);
 
-   return pt;
-}
-
-
-/**
- * Draw quad with texcoords and optional color.
- * Coords are gallium window coords with y=0=top.
- * \param color  may be null
- * \param invertTex  if true, flip texcoords vertically
- */
-static void
-draw_quad(struct gl_context *ctx, GLfloat x0, GLfloat y0, GLfloat z,
-          GLfloat x1, GLfloat y1, const GLfloat *color,
-          GLboolean invertTex, GLfloat maxXcoord, GLfloat maxYcoord)
-{
-   struct st_context *st = st_context(ctx);
-   struct pipe_context *pipe = st->pipe;
-   GLfloat (*verts)[3][4]; /* four verts, three attribs, XYZW */
-   struct pipe_resource *buf = NULL;
-   unsigned offset;
-
-   u_upload_alloc(st->uploader, 0, 4 * sizeof(verts[0]), 4, &offset,
-                  &buf, (void **) &verts);
-   if (!buf) {
-      return;
-   }
-
-   /* setup vertex data */
-   {
-      const struct gl_framebuffer *fb = st->ctx->DrawBuffer;
-      const GLfloat fb_width = (GLfloat) fb->Width;
-      const GLfloat fb_height = (GLfloat) fb->Height;
-      const GLfloat clip_x0 = x0 / fb_width * 2.0f - 1.0f;
-      const GLfloat clip_y0 = y0 / fb_height * 2.0f - 1.0f;
-      const GLfloat clip_x1 = x1 / fb_width * 2.0f - 1.0f;
-      const GLfloat clip_y1 = y1 / fb_height * 2.0f - 1.0f;
-      const GLfloat sLeft = 0.0f, sRight = maxXcoord;
-      const GLfloat tTop = invertTex ? maxYcoord : 0.0f;
-      const GLfloat tBot = invertTex ? 0.0f : maxYcoord;
-      GLuint i;
-
-      /* upper-left */
-      verts[0][0][0] = clip_x0;    /* v[0].attr[0].x */
-      verts[0][0][1] = clip_y0;    /* v[0].attr[0].y */
-
-      /* upper-right */
-      verts[1][0][0] = clip_x1;
-      verts[1][0][1] = clip_y0;
-
-      /* lower-right */
-      verts[2][0][0] = clip_x1;
-      verts[2][0][1] = clip_y1;
-
-      /* lower-left */
-      verts[3][0][0] = clip_x0;
-      verts[3][0][1] = clip_y1;
-
-      verts[0][1][0] = sLeft; /* v[0].attr[1].S */
-      verts[0][1][1] = tTop;  /* v[0].attr[1].T */
-      verts[1][1][0] = sRight;
-      verts[1][1][1] = tTop;
-      verts[2][1][0] = sRight;
-      verts[2][1][1] = tBot;
-      verts[3][1][0] = sLeft;
-      verts[3][1][1] = tBot;
-
-      /* same for all verts: */
-      if (color) {
-         for (i = 0; i < 4; i++) {
-            verts[i][0][2] = z;         /* v[i].attr[0].z */
-            verts[i][0][3] = 1.0f;      /* v[i].attr[0].w */
-            verts[i][2][0] = color[0];  /* v[i].attr[2].r */
-            verts[i][2][1] = color[1];  /* v[i].attr[2].g */
-            verts[i][2][2] = color[2];  /* v[i].attr[2].b */
-            verts[i][2][3] = color[3];  /* v[i].attr[2].a */
-            verts[i][1][2] = 0.0f;      /* v[i].attr[1].R */
-            verts[i][1][3] = 1.0f;      /* v[i].attr[1].Q */
-         }
-      }
-      else {
-         for (i = 0; i < 4; i++) {
-            verts[i][0][2] = z;    /*Z*/
-            verts[i][0][3] = 1.0f; /*W*/
-            verts[i][1][2] = 0.0f; /*R*/
-            verts[i][1][3] = 1.0f; /*Q*/
-         }
-      }
-   }
+#if USE_DRAWPIXELS_CACHE
+   cache_drawpixels_image(st, width, height, format, type, unpack, pixels, pt);
+#endif
 
-   u_upload_unmap(st->uploader);
-   util_draw_vertex_buffer(pipe, st->cso_context, buf,
-                           cso_get_aux_vertex_buffer_slot(st->cso_context),
-                           offset,
-                          PIPE_PRIM_QUADS,
-                          4,  /* verts */
-                          3); /* attribs/vert */
-   pipe_resource_reference(&buf, NULL);
+   return pt;
 }
 
 
-
 static void
 draw_textured_quad(struct gl_context *ctx, GLint x, GLint y, GLfloat z,
                    GLsizei width, GLsizei height,
@@ -554,35 +748,37 @@ draw_textured_quad(struct gl_context *ctx, GLint x, GLint y, GLfloat z,
    struct st_context *st = st_context(ctx);
    struct pipe_context *pipe = st->pipe;
    struct cso_context *cso = st->cso_context;
+   const unsigned fb_width = _mesa_geometric_width(ctx->DrawBuffer);
+   const unsigned fb_height = _mesa_geometric_height(ctx->DrawBuffer);
    GLfloat x0, y0, x1, y1;
-   GLsizei maxSize;
-   boolean normalized = sv[0]->texture->target != PIPE_TEXTURE_RECT;
+   ASSERTED GLsizei maxSize;
+   boolean normalized = sv[0]->texture->target == PIPE_TEXTURE_2D;
+   unsigned cso_state_mask;
+
+   assert(sv[0]->texture->target == st->internal_target);
 
    /* limit checks */
    /* XXX if DrawPixels image is larger than max texture size, break
     * it up into chunks.
     */
-   maxSize = 1 << (pipe->screen->get_param(pipe->screen,
-                                        PIPE_CAP_MAX_TEXTURE_2D_LEVELS) - 1);
+   maxSize = pipe->screen->get_param(pipe->screen,
+                                     PIPE_CAP_MAX_TEXTURE_2D_SIZE);
    assert(width <= maxSize);
    assert(height <= maxSize);
 
-   cso_save_rasterizer(cso);
-   cso_save_viewport(cso);
-   cso_save_fragment_samplers(cso);
-   cso_save_fragment_sampler_views(cso);
-   cso_save_fragment_shader(cso);
-   cso_save_stream_outputs(cso);
-   cso_save_vertex_shader(cso);
-   cso_save_tessctrl_shader(cso);
-   cso_save_tesseval_shader(cso);
-   cso_save_geometry_shader(cso);
-   cso_save_vertex_elements(cso);
-   cso_save_aux_vertex_buffer_slot(cso);
+   cso_state_mask = (CSO_BIT_RASTERIZER |
+                     CSO_BIT_VIEWPORT |
+                     CSO_BIT_FRAGMENT_SAMPLERS |
+                     CSO_BIT_FRAGMENT_SAMPLER_VIEWS |
+                     CSO_BIT_STREAM_OUTPUTS |
+                     CSO_BIT_VERTEX_ELEMENTS |
+                     CSO_BIT_AUX_VERTEX_BUFFER_SLOT |
+                     CSO_BITS_ALL_SHADERS);
    if (write_stencil) {
-      cso_save_depth_stencil_alpha(cso);
-      cso_save_blend(cso);
+      cso_state_mask |= (CSO_BIT_DEPTH_STENCIL_ALPHA |
+                         CSO_BIT_BLEND);
    }
+   cso_save_state(cso, cso_state_mask);
 
    /* rasterizer state: just scissor */
    {
@@ -592,7 +788,10 @@ draw_textured_quad(struct gl_context *ctx, GLint x, GLint y, GLfloat z,
                                         ctx->Color._ClampFragmentColor;
       rasterizer.half_pixel_center = 1;
       rasterizer.bottom_edge_rule = 1;
-      rasterizer.depth_clip = !ctx->Transform.DepthClamp;
+      rasterizer.depth_clip_near = st->clamp_frag_depth_in_shader ||
+                                   !ctx->Transform.DepthClampNear;
+      rasterizer.depth_clip_far = st->clamp_frag_depth_in_shader ||
+                                  !ctx->Transform.DepthClampFar;
       rasterizer.scissor = ctx->Scissor.EnableFlags;
       cso_set_rasterizer(cso, &rasterizer);
    }
@@ -648,13 +847,15 @@ draw_textured_quad(struct gl_context *ctx, GLint x, GLint y, GLfloat z,
       sampler.normalized_coords = normalized;
 
       if (fpv) {
+         /* drawing a color image */
          const struct pipe_sampler_state *samplers[PIPE_MAX_SAMPLERS];
-         uint num = MAX2(MAX2(fpv->drawpix_sampler, fpv->pixelmap_sampler) + 1,
-                         st->state.num_samplers[PIPE_SHADER_FRAGMENT]);
+         uint num = MAX3(fpv->drawpix_sampler + 1,
+                         fpv->pixelmap_sampler + 1,
+                         st->state.num_frag_samplers);
          uint i;
 
-         for (i = 0; i < st->state.num_samplers[PIPE_SHADER_FRAGMENT]; i++)
-            samplers[i] = &st->state.samplers[PIPE_SHADER_FRAGMENT][i];
+         for (i = 0; i < st->state.num_frag_samplers; i++)
+            samplers[i] = &st->state.frag_samplers[i];
 
          samplers[fpv->drawpix_sampler] = &sampler;
          if (sv[1])
@@ -662,52 +863,45 @@ draw_textured_quad(struct gl_context *ctx, GLint x, GLint y, GLfloat z,
 
          cso_set_samplers(cso, PIPE_SHADER_FRAGMENT, num, samplers);
       } else {
+         /* drawing a depth/stencil image */
          const struct pipe_sampler_state *samplers[2] = {&sampler, &sampler};
 
          cso_set_samplers(cso, PIPE_SHADER_FRAGMENT, num_sampler_view, samplers);
       }
    }
 
-   /* viewport state: viewport matching window dims */
-   {
-      const float w = (float) ctx->DrawBuffer->Width;
-      const float h = (float) ctx->DrawBuffer->Height;
-      struct pipe_viewport_state vp;
-      vp.scale[0] =  0.5f * w;
-      vp.scale[1] = -0.5f * h;
-      vp.scale[2] = 0.5f;
-      vp.translate[0] = 0.5f * w;
-      vp.translate[1] = 0.5f * h;
-      vp.translate[2] = 0.5f;
-      cso_set_viewport(cso, &vp);
-   }
-
-   cso_set_vertex_elements(cso, 3, st->velems_util_draw);
-   cso_set_stream_outputs(st->cso_context, 0, NULL, NULL);
-
    /* user textures, plus the drawpix textures */
    if (fpv) {
+      /* drawing a color image */
       struct pipe_sampler_view *sampler_views[PIPE_MAX_SAMPLERS];
       uint num = MAX3(fpv->drawpix_sampler + 1,
                       fpv->pixelmap_sampler + 1,
                       st->state.num_sampler_views[PIPE_SHADER_FRAGMENT]);
 
-      memcpy(sampler_views, st->state.sampler_views[PIPE_SHADER_FRAGMENT],
+      memcpy(sampler_views, st->state.frag_sampler_views,
              sizeof(sampler_views));
 
       sampler_views[fpv->drawpix_sampler] = sv[0];
       if (sv[1])
          sampler_views[fpv->pixelmap_sampler] = sv[1];
       cso_set_sampler_views(cso, PIPE_SHADER_FRAGMENT, num, sampler_views);
-   } else
+   } else {
+      /* drawing a depth/stencil image */
       cso_set_sampler_views(cso, PIPE_SHADER_FRAGMENT, num_sampler_view, sv);
+   }
+
+   /* viewport state: viewport matching window dims */
+   cso_set_viewport_dims(cso, fb_width, fb_height, TRUE);
+
+   cso_set_vertex_elements(cso, 3, st->util_velems);
+   cso_set_stream_outputs(cso, 0, NULL, NULL);
 
    /* Compute Gallium window coords (y=0=top) with pixel zoom.
     * Recall that these coords are transformed by the current
     * vertex shader and viewport transformation.
     */
    if (st_fb_orientation(ctx->DrawBuffer) == Y_0_BOTTOM) {
-      y = ctx->DrawBuffer->Height - (int) (y + height * ctx->Pixel.ZoomY);
+      y = fb_height - (int) (y + height * ctx->Pixel.ZoomY);
       invertTex = !invertTex;
    }
 
@@ -719,27 +913,27 @@ draw_textured_quad(struct gl_context *ctx, GLint x, GLint y, GLfloat z,
    /* convert Z from [0,1] to [-1,-1] to match viewport Z scale/bias */
    z = z * 2.0f - 1.0f;
 
-   draw_quad(ctx, x0, y0, z, x1, y1, color, invertTex,
-             normalized ? ((GLfloat) width / sv[0]->texture->width0) : (GLfloat)width,
-             normalized ? ((GLfloat) height / sv[0]->texture->height0) : (GLfloat)height);
+   {
+      const float clip_x0 = x0 / (float) fb_width * 2.0f - 1.0f;
+      const float clip_y0 = y0 / (float) fb_height * 2.0f - 1.0f;
+      const float clip_x1 = x1 / (float) fb_width * 2.0f - 1.0f;
+      const float clip_y1 = y1 / (float) fb_height * 2.0f - 1.0f;
+      const float maxXcoord = normalized ?
+         ((float) width / sv[0]->texture->width0) : (float) width;
+      const float maxYcoord = normalized
+         ? ((float) height / sv[0]->texture->height0) : (float) height;
+      const float sLeft = 0.0f, sRight = maxXcoord;
+      const float tTop = invertTex ? maxYcoord : 0.0f;
+      const float tBot = invertTex ? 0.0f : maxYcoord;
+
+      if (!st_draw_quad(st, clip_x0, clip_y0, clip_x1, clip_y1, z,
+                        sLeft, tBot, sRight, tTop, color, 0)) {
+         _mesa_error(ctx, GL_OUT_OF_MEMORY, "glDrawPixels");
+      }
+   }
 
    /* restore state */
-   cso_restore_rasterizer(cso);
-   cso_restore_viewport(cso);
-   cso_restore_fragment_samplers(cso);
-   cso_restore_fragment_sampler_views(cso);
-   cso_restore_fragment_shader(cso);
-   cso_restore_vertex_shader(cso);
-   cso_restore_tessctrl_shader(cso);
-   cso_restore_tesseval_shader(cso);
-   cso_restore_geometry_shader(cso);
-   cso_restore_vertex_elements(cso);
-   cso_restore_aux_vertex_buffer_slot(cso);
-   cso_restore_stream_outputs(cso);
-   if (write_stencil) {
-      cso_restore_depth_stencil_alpha(cso);
-      cso_restore_blend(cso);
-   }
+   cso_restore_state(cso);
 }
 
 
@@ -751,7 +945,7 @@ static void
 draw_stencil_pixels(struct gl_context *ctx, GLint x, GLint y,
                     GLsizei width, GLsizei height, GLenum format, GLenum type,
                     const struct gl_pixelstore_attrib *unpack,
-                    const GLvoid *pixels)
+                    const void *pixels)
 {
    struct st_context *st = st_context(ctx);
    struct pipe_context *pipe = st->pipe;
@@ -805,7 +999,7 @@ draw_stencil_pixels(struct gl_context *ctx, GLint x, GLint y,
       for (row = 0; row < height; row++) {
          GLfloat *zValuesFloat = (GLfloat*)zValues;
          GLenum destType = GL_UNSIGNED_BYTE;
-         const GLvoid *source = _mesa_image_address2d(&clippedUnpack, pixels,
+         const void *source = _mesa_image_address2d(&clippedUnpack, pixels,
                                                       width, height,
                                                       format, type,
                                                       row, 0);
@@ -948,7 +1142,36 @@ get_color_fp_variant(struct st_context *st)
                        ctx->Pixel.AlphaScale != 1.0);
    key.pixelMaps = ctx->Pixel.MapColorFlag;
    key.clamp_color = st->clamp_frag_color_in_shader &&
-                     st->ctx->Color._ClampFragmentColor;
+                     ctx->Color._ClampFragmentColor;
+
+   fpv = st_get_fp_variant(st, st->fp, &key);
+
+   return fpv;
+}
+
+/**
+ * Get fragment program variant for a glDrawPixels command
+ * for COLOR_INDEX data
+ */
+static struct st_fp_variant *
+get_color_index_fp_variant(struct st_context *st)
+{
+   struct gl_context *ctx = st->ctx;
+   struct st_fp_variant_key key;
+   struct st_fp_variant *fpv;
+
+   memset(&key, 0, sizeof(key));
+
+   key.st = st->has_shareable_shaders ? NULL : st;
+   key.drawpixels = 1;
+   /* Since GL is always in RGBA mode MapColorFlag does not
+    * affect GL_COLOR_INDEX format.
+    * Scale and bias also never affect GL_COLOR_INDEX format.
+    */
+   key.scaleAndBias = 0;
+   key.pixelMaps = 0;
+   key.clamp_color = st->clamp_frag_color_in_shader &&
+                     ctx->Color._ClampFragmentColor;
 
    fpv = st_get_fp_variant(st, st->fp, &key);
 
@@ -963,9 +1186,8 @@ static void
 clamp_size(struct pipe_context *pipe, GLsizei *width, GLsizei *height,
            struct gl_pixelstore_attrib *unpack)
 {
-   const int maxSize =
-      1 << (pipe->screen->get_param(pipe->screen,
-                                    PIPE_CAP_MAX_TEXTURE_2D_LEVELS) - 1);
+   const int maxSize = pipe->screen->get_param(pipe->screen,
+                                               PIPE_CAP_MAX_TEXTURE_2D_SIZE);
 
    if (*width > maxSize) {
       if (unpack->RowLength == 0)
@@ -1018,17 +1240,17 @@ setup_sampler_swizzle(struct pipe_sampler_view *sv, GLenum format, GLenum type)
 
       /* invert the format's swizzle to setup the sampler's swizzle */
       if (format == GL_RGBA) {
-         c0 = UTIL_FORMAT_SWIZZLE_X;
-         c1 = UTIL_FORMAT_SWIZZLE_Y;
-         c2 = UTIL_FORMAT_SWIZZLE_Z;
-         c3 = UTIL_FORMAT_SWIZZLE_W;
+         c0 = PIPE_SWIZZLE_X;
+         c1 = PIPE_SWIZZLE_Y;
+         c2 = PIPE_SWIZZLE_Z;
+         c3 = PIPE_SWIZZLE_W;
       }
       else {
          assert(format == GL_BGRA);
-         c0 = UTIL_FORMAT_SWIZZLE_Z;
-         c1 = UTIL_FORMAT_SWIZZLE_Y;
-         c2 = UTIL_FORMAT_SWIZZLE_X;
-         c3 = UTIL_FORMAT_SWIZZLE_W;
+         c0 = PIPE_SWIZZLE_Z;
+         c1 = PIPE_SWIZZLE_Y;
+         c2 = PIPE_SWIZZLE_X;
+         c3 = PIPE_SWIZZLE_W;
       }
       sv->swizzle_r = search_swizzle(desc->swizzle, c0);
       sv->swizzle_g = search_swizzle(desc->swizzle, c1);
@@ -1041,6 +1263,34 @@ setup_sampler_swizzle(struct pipe_sampler_view *sv, GLenum format, GLenum type)
 }
 
 
+/**
+ * Compute the effective raster z position. This performs depth-clamping
+ * if needed.
+ */
+static float
+get_effective_raster_z(struct gl_context *ctx)
+{
+   float z = ctx->Current.RasterPos[2];
+   if (st_context(ctx)->clamp_frag_depth_in_shader) {
+      GLfloat depth_near;
+      GLfloat depth_far;
+      if (ctx->ViewportArray[0].Near < ctx->ViewportArray[0].Far) {
+         depth_near = ctx->ViewportArray[0].Near;
+         depth_far = ctx->ViewportArray[0].Far;
+      } else {
+         depth_near = ctx->ViewportArray[0].Far;
+         depth_far = ctx->ViewportArray[0].Near;
+      }
+
+      if (ctx->Transform.DepthClampNear)
+         z = MAX2(z, depth_near);
+      if (ctx->Transform.DepthClampFar)
+         z = MIN2(z, depth_far);
+   }
+   return z;
+}
+
+
 /**
  * Called via ctx->Driver.DrawPixels()
  */
@@ -1048,11 +1298,10 @@ static void
 st_DrawPixels(struct gl_context *ctx, GLint x, GLint y,
               GLsizei width, GLsizei height,
               GLenum format, GLenum type,
-              const struct gl_pixelstore_attrib *unpack, const GLvoid *pixels)
+              const struct gl_pixelstore_attrib *unpack, const void *pixels)
 {
-   void *driver_vp, *driver_fp;
+   void *driver_fp;
    struct st_context *st = st_context(ctx);
-   const GLfloat *color;
    struct pipe_context *pipe = st->pipe;
    GLboolean write_stencil = GL_FALSE, write_depth = GL_FALSE;
    struct pipe_sampler_view *sv[2] = { NULL };
@@ -1064,9 +1313,12 @@ st_DrawPixels(struct gl_context *ctx, GLint x, GLint y,
    /* Mesa state should be up to date by now */
    assert(ctx->NewState == 0x0);
 
+   _mesa_update_draw_buffer_bounds(ctx, ctx->DrawBuffer);
+
    st_flush_bitmap_cache(st);
+   st_invalidate_readpix_cache(st);
 
-   st_validate_state(st);
+   st_validate_state(st, ST_PIPELINE_META);
 
    /* Limit the size of the glDrawPixels to the max texture size.
     * Strictly speaking, that's not correct but since we don't handle
@@ -1091,23 +1343,29 @@ st_DrawPixels(struct gl_context *ctx, GLint x, GLint y,
       return;
    }
 
+   /* Put glDrawPixels image into a texture */
+   pt = make_texture(st, width, height, format, type, unpack, pixels);
+   if (!pt) {
+      _mesa_error(ctx, GL_OUT_OF_MEMORY, "glDrawPixels");
+      return;
+   }
+
+   st_make_passthrough_vertex_shader(st);
+
    /*
     * Get vertex/fragment shaders
     */
    if (write_depth || write_stencil) {
       driver_fp = get_drawpix_z_stencil_program(st, write_depth,
                                                 write_stencil);
-      driver_vp = make_passthrough_vertex_shader(st, GL_TRUE);
-      color = ctx->Current.RasterColor;
    }
    else {
-      fpv = get_color_fp_variant(st);
+      fpv = (format != GL_COLOR_INDEX) ? get_color_fp_variant(st) :
+                                         get_color_index_fp_variant(st);
 
       driver_fp = fpv->driver_shader;
-      driver_vp = make_passthrough_vertex_shader(st, GL_FALSE);
 
-      color = NULL;
-      if (ctx->Pixel.MapColorFlag) {
+      if (ctx->Pixel.MapColorFlag && format != GL_COLOR_INDEX) {
          pipe_sampler_view_reference(&sv[1],
                                      st->pixel_xfer.pixelmap_sampler_view);
          num_sampler_view++;
@@ -1116,15 +1374,7 @@ st_DrawPixels(struct gl_context *ctx, GLint x, GLint y,
       /* compiling a new fragment shader variant added new state constants
        * into the constant buffer, we need to update them
        */
-      st_upload_constants(st, st->fp->Base.Base.Parameters,
-                          PIPE_SHADER_FRAGMENT);
-   }
-
-   /* Put glDrawPixels image into a texture */
-   pt = make_texture(st, width, height, format, type, unpack, pixels);
-   if (!pt) {
-      _mesa_error(ctx, GL_OUT_OF_MEMORY, "glDrawPixels");
-      return;
+      st_upload_constants(st, &st->fp->Base);
    }
 
    /* create sampler view for the image */
@@ -1157,18 +1407,20 @@ st_DrawPixels(struct gl_context *ctx, GLint x, GLint y,
       num_sampler_view++;
    }
 
-   draw_textured_quad(ctx, x, y, ctx->Current.RasterPos[2],
+   draw_textured_quad(ctx, x, y, get_effective_raster_z(ctx),
                       width, height,
                       ctx->Pixel.ZoomX, ctx->Pixel.ZoomY,
                       sv,
                       num_sampler_view,
-                      driver_vp,
+                      st->passthrough_vs,
                       driver_fp, fpv,
-                      color, GL_FALSE, write_depth, write_stencil);
+                      ctx->Current.RasterColor,
+                      GL_FALSE, write_depth, write_stencil);
    pipe_sampler_view_reference(&sv[0], NULL);
    if (num_sampler_view > 1)
       pipe_sampler_view_reference(&sv[1], NULL);
 
+   /* free the texture (but may persist in the cache) */
    pipe_resource_reference(&pt, NULL);
 }
 
@@ -1296,20 +1548,21 @@ blit_copy_pixels(struct gl_context *ctx, GLint srcx, GLint srcy,
    struct gl_pixelstore_attrib pack, unpack;
    GLint readX, readY, readW, readH, drawX, drawY, drawW, drawH;
 
-   if (type == GL_COLOR &&
-       ctx->Pixel.ZoomX == 1.0 &&
+   if (ctx->Pixel.ZoomX == 1.0 &&
        ctx->Pixel.ZoomY == 1.0 &&
-       ctx->_ImageTransferState == 0x0 &&
-       !ctx->Color.BlendEnabled &&
-       !ctx->Color.AlphaEnabled &&
-       (!ctx->Color.ColorLogicOpEnabled || ctx->Color.LogicOp == GL_COPY) &&
-       !ctx->Depth.Test &&
-       !ctx->Fog.Enabled &&
-       !ctx->Stencil.Enabled &&
-       !ctx->FragmentProgram.Enabled &&
-       !ctx->VertexProgram.Enabled &&
-       !ctx->_Shader->CurrentProgram[MESA_SHADER_FRAGMENT] &&
-       ctx->DrawBuffer->_NumColorDrawBuffers == 1 &&
+       (type != GL_COLOR ||
+        (ctx->_ImageTransferState == 0x0 &&
+         !ctx->Color.BlendEnabled &&
+         !ctx->Color.AlphaEnabled &&
+         (!ctx->Color.ColorLogicOpEnabled || ctx->Color.LogicOp == GL_COPY) &&
+         !ctx->Depth.Test &&
+         !ctx->Fog.Enabled &&
+         !ctx->Stencil.Enabled &&
+         !ctx->FragmentProgram.Enabled &&
+         !ctx->VertexProgram.Enabled &&
+         !ctx->_Shader->CurrentProgram[MESA_SHADER_FRAGMENT] &&
+         !_mesa_ati_fragment_shader_enabled(ctx) &&
+         ctx->DrawBuffer->_NumColorDrawBuffers == 1)) &&
        !ctx->Query.CondRenderQuery &&
        !ctx->Query.CurrentOcclusionObject) {
       struct st_renderbuffer *rbRead, *rbDraw;
@@ -1342,8 +1595,18 @@ blit_copy_pixels(struct gl_context *ctx, GLint srcx, GLint srcy,
       drawW = readW;
       drawH = readH;
 
-      rbRead = st_get_color_read_renderbuffer(ctx);
-      rbDraw = st_renderbuffer(ctx->DrawBuffer->_ColorDrawBuffers[0]);
+      if (type == GL_COLOR) {
+         rbRead = st_get_color_read_renderbuffer(ctx);
+         rbDraw = st_renderbuffer(ctx->DrawBuffer->_ColorDrawBuffers[0]);
+      } else if (type == GL_DEPTH || type == GL_DEPTH_STENCIL) {
+         rbRead = st_renderbuffer(ctx->ReadBuffer->Attachment[BUFFER_DEPTH].Renderbuffer);
+         rbDraw = st_renderbuffer(ctx->DrawBuffer->Attachment[BUFFER_DEPTH].Renderbuffer);
+      } else if (type == GL_STENCIL) {
+         rbRead = st_renderbuffer(ctx->ReadBuffer->Attachment[BUFFER_STENCIL].Renderbuffer);
+         rbDraw = st_renderbuffer(ctx->DrawBuffer->Attachment[BUFFER_STENCIL].Renderbuffer);
+      } else {
+         return false;
+      }
 
       /* Flip src/dst position depending on the orientation of buffers. */
       if (st_fb_orientation(ctx->ReadBuffer) == Y_0_TOP) {
@@ -1384,16 +1647,29 @@ blit_copy_pixels(struct gl_context *ctx, GLint srcx, GLint srcy,
          blit.dst.box.width = drawW;
          blit.dst.box.height = drawH;
          blit.dst.box.depth = 1;
-         blit.mask = PIPE_MASK_RGBA;
          blit.filter = PIPE_TEX_FILTER_NEAREST;
 
+         if (type == GL_COLOR)
+            blit.mask |= PIPE_MASK_RGBA;
+         if (type == GL_DEPTH)
+            blit.mask |= PIPE_MASK_Z;
+         if (type == GL_STENCIL)
+            blit.mask |= PIPE_MASK_S;
+         if (type == GL_DEPTH_STENCIL)
+            blit.mask |= PIPE_MASK_ZS;
+
+         if (ctx->DrawBuffer != ctx->WinSysDrawBuffer)
+            st_window_rectangles_to_blit(ctx, &blit);
+
          if (screen->is_format_supported(screen, blit.src.format,
                                          blit.src.resource->target,
                                          blit.src.resource->nr_samples,
+                                         blit.src.resource->nr_storage_samples,
                                          PIPE_BIND_SAMPLER_VIEW) &&
              screen->is_format_supported(screen, blit.dst.format,
                                          blit.dst.resource->target,
                                          blit.dst.resource->nr_samples,
+                                         blit.dst.resource->nr_storage_samples,
                                          PIPE_BIND_RENDER_TARGET)) {
             pipe->blit(pipe, &blit);
             return GL_TRUE;
@@ -1414,21 +1690,26 @@ st_CopyPixels(struct gl_context *ctx, GLint srcx, GLint srcy,
    struct pipe_context *pipe = st->pipe;
    struct pipe_screen *screen = pipe->screen;
    struct st_renderbuffer *rbRead;
-   void *driver_vp, *driver_fp;
+   void *driver_fp;
    struct pipe_resource *pt;
    struct pipe_sampler_view *sv[2] = { NULL };
    struct st_fp_variant *fpv = NULL;
    int num_sampler_view = 1;
-   GLfloat *color;
    enum pipe_format srcFormat;
    unsigned srcBind;
    GLboolean invertTex = GL_FALSE;
    GLint readX, readY, readW, readH;
    struct gl_pixelstore_attrib pack = ctx->DefaultPacking;
 
+   _mesa_update_draw_buffer_bounds(ctx, ctx->DrawBuffer);
+
    st_flush_bitmap_cache(st);
+   st_invalidate_readpix_cache(st);
 
-   st_validate_state(st);
+   st_validate_state(st, ST_PIPELINE_META);
+
+   if (blit_copy_pixels(ctx, srcx, srcy, width, height, dstx, dsty, type))
+      return;
 
    if (type == GL_DEPTH_STENCIL) {
       /* XXX make this more efficient */
@@ -1443,9 +1724,6 @@ st_CopyPixels(struct gl_context *ctx, GLint srcx, GLint srcy,
       return;
    }
 
-   if (blit_copy_pixels(ctx, srcx, srcy, width, height, dstx, dsty, type))
-      return;
-
    /*
     * The subsequent code implements glCopyPixels by copying the source
     * pixels into a temporary texture that's then applied to a textured quad.
@@ -1453,6 +1731,7 @@ st_CopyPixels(struct gl_context *ctx, GLint srcx, GLint srcy,
     * are handled.
     */
 
+   st_make_passthrough_vertex_shader(st);
 
    /*
     * Get vertex/fragment shaders
@@ -1461,10 +1740,8 @@ st_CopyPixels(struct gl_context *ctx, GLint srcx, GLint srcy,
       fpv = get_color_fp_variant(st);
 
       rbRead = st_get_color_read_renderbuffer(ctx);
-      color = NULL;
 
       driver_fp = fpv->driver_shader;
-      driver_vp = make_passthrough_vertex_shader(st, GL_FALSE);
 
       if (ctx->Pixel.MapColorFlag) {
          pipe_sampler_view_reference(&sv[1],
@@ -1475,17 +1752,14 @@ st_CopyPixels(struct gl_context *ctx, GLint srcx, GLint srcy,
       /* compiling a new fragment shader variant added new state constants
        * into the constant buffer, we need to update them
        */
-      st_upload_constants(st, st->fp->Base.Base.Parameters,
-                          PIPE_SHADER_FRAGMENT);
+      st_upload_constants(st, &st->fp->Base);
    }
    else {
       assert(type == GL_DEPTH);
       rbRead = st_renderbuffer(ctx->ReadBuffer->
                                Attachment[BUFFER_DEPTH].Renderbuffer);
-      color = ctx->Current.Attrib[VERT_ATTRIB_COLOR0];
 
       driver_fp = get_drawpix_z_stencil_program(st, GL_TRUE, GL_FALSE);
-      driver_vp = make_passthrough_vertex_shader(st, GL_TRUE);
    }
 
    /* Choose the format for the temporary texture. */
@@ -1494,40 +1768,40 @@ st_CopyPixels(struct gl_context *ctx, GLint srcx, GLint srcy,
       (type == GL_COLOR ? PIPE_BIND_RENDER_TARGET : PIPE_BIND_DEPTH_STENCIL);
 
    if (!screen->is_format_supported(screen, srcFormat, st->internal_target, 0,
-                                    srcBind)) {
+                                    0, srcBind)) {
       /* srcFormat is non-renderable. Find a compatible renderable format. */
       if (type == GL_DEPTH) {
          srcFormat = st_choose_format(st, GL_DEPTH_COMPONENT, GL_NONE,
-                                      GL_NONE, st->internal_target, 0,
-                                      srcBind, FALSE);
+                                      GL_NONE, st->internal_target, 0, 0,
+                                      srcBind, false, false);
       }
       else {
          assert(type == GL_COLOR);
 
          if (util_format_is_float(srcFormat)) {
             srcFormat = st_choose_format(st, GL_RGBA32F, GL_NONE,
-                                         GL_NONE, st->internal_target, 0,
-                                         srcBind, FALSE);
+                                         GL_NONE, st->internal_target, 0, 0,
+                                         srcBind, false, false);
          }
          else if (util_format_is_pure_sint(srcFormat)) {
             srcFormat = st_choose_format(st, GL_RGBA32I, GL_NONE,
-                                         GL_NONE, st->internal_target, 0,
-                                         srcBind, FALSE);
+                                         GL_NONE, st->internal_target, 0, 0,
+                                         srcBind, false, false);
          }
          else if (util_format_is_pure_uint(srcFormat)) {
             srcFormat = st_choose_format(st, GL_RGBA32UI, GL_NONE,
-                                         GL_NONE, st->internal_target, 0,
-                                         srcBind, FALSE);
+                                         GL_NONE, st->internal_target, 0, 0,
+                                         srcBind, false, false);
          }
          else if (util_format_is_snorm(srcFormat)) {
             srcFormat = st_choose_format(st, GL_RGBA16_SNORM, GL_NONE,
-                                         GL_NONE, st->internal_target, 0,
-                                         srcBind, FALSE);
+                                         GL_NONE, st->internal_target, 0, 0,
+                                         srcBind, false, false);
          }
          else {
             srcFormat = st_choose_format(st, GL_RGBA, GL_NONE,
-                                         GL_NONE, st->internal_target, 0,
-                                         srcBind, FALSE);
+                                         GL_NONE, st->internal_target, 0, 0,
+                                         srcBind, false, false);
          }
       }
 
@@ -1608,13 +1882,15 @@ st_CopyPixels(struct gl_context *ctx, GLint srcx, GLint srcy,
    /* OK, the texture 'pt' contains the src image/pixels.  Now draw a
     * textured quad with that texture.
     */
-   draw_textured_quad(ctx, dstx, dsty, ctx->Current.RasterPos[2],
+
+   draw_textured_quad(ctx, dstx, dsty, get_effective_raster_z(ctx),
                       width, height, ctx->Pixel.ZoomX, ctx->Pixel.ZoomY,
                       sv,
                       num_sampler_view,
-                      driver_vp, 
+                      st->passthrough_vs,
                       driver_fp, fpv,
-                      color, invertTex, GL_FALSE, GL_FALSE);
+                      ctx->Current.Attrib[VERT_ATTRIB_COLOR0],
+                      invertTex, GL_FALSE, GL_FALSE);
 
    pipe_resource_reference(&pt, NULL);
    pipe_sampler_view_reference(&sv[0], NULL);
@@ -1640,8 +1916,13 @@ st_destroy_drawpix(struct st_context *st)
                                     st->drawpix.zs_shaders[i]);
    }
 
-   if (st->drawpix.vert_shaders[0])
-      cso_delete_vertex_shader(st->cso_context, st->drawpix.vert_shaders[0]);
-   if (st->drawpix.vert_shaders[1])
-      cso_delete_vertex_shader(st->cso_context, st->drawpix.vert_shaders[1]);
+   if (st->passthrough_vs)
+      cso_delete_vertex_shader(st->cso_context, st->passthrough_vs);
+
+   /* Free cache data */
+   for (i = 0; i < ARRAY_SIZE(st->drawpix_cache.entries); i++) {
+      struct drawpix_cache_entry *entry = &st->drawpix_cache.entries[i];
+      free(entry->image);
+      pipe_resource_reference(&entry->texture, NULL);
+   }
 }