mesa: enable msaa in clear_with_quad if needed
[mesa.git] / src / mesa / state_tracker / st_cb_clear.c
index a5aa8f496028ed2ceb9b07b17998057d97ad872c..1e28e27586a81932a305d4a5785639f51aa75393 100644 (file)
@@ -1,6 +1,6 @@
 /**************************************************************************
  * 
- * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * Copyright 2007 VMware, Inc.
  * All Rights Reserved.
  * Copyright 2009 VMware, Inc.  All Rights Reserved.
  * 
@@ -19,7 +19,7 @@
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
- * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
  /*
   * Authors:
-  *   Keith Whitwell <keith@tungstengraphics.com>
+  *   Keith Whitwell <keithw@vmware.com>
   *   Brian Paul
   *   Michel Dänzer
   */
 
+#include "main/errors.h"
 #include "main/glheader.h"
 #include "main/accum.h"
 #include "main/formats.h"
+#include "main/framebuffer.h"
 #include "main/macros.h"
 #include "main/glformats.h"
 #include "program/prog_instruction.h"
 #include "st_context.h"
 #include "st_atom.h"
+#include "st_cb_bitmap.h"
 #include "st_cb_clear.h"
 #include "st_cb_fbo.h"
+#include "st_draw.h"
 #include "st_format.h"
+#include "st_nir.h"
 #include "st_program.h"
+#include "st_util.h"
 
 #include "pipe/p_context.h"
 #include "pipe/p_shader_tokens.h"
@@ -53,8 +59,6 @@
 #include "util/u_format.h"
 #include "util/u_inlines.h"
 #include "util/u_simple_shaders.h"
-#include "util/u_draw_quad.h"
-#include "util/u_upload_mgr.h"
 
 #include "cso_cache/cso_context.h"
 
@@ -67,8 +71,10 @@ st_init_clear(struct st_context *st)
 {
    memset(&st->clear, 0, sizeof(st->clear));
 
-   st->clear.raster.gl_rasterization_rules = 1;
-   st->clear.raster.depth_clip = 1;
+   st->clear.raster.half_pixel_center = 1;
+   st->clear.raster.bottom_edge_rule = 1;
+   st->clear.raster.depth_clip_near = 1;
+   st->clear.raster.depth_clip_far = 1;
 }
 
 
@@ -86,127 +92,164 @@ st_destroy_clear(struct st_context *st)
       cso_delete_vertex_shader(st->cso_context, st->clear.vs);
       st->clear.vs = NULL;
    }
+   if (st->clear.vs_layered) {
+      cso_delete_vertex_shader(st->cso_context, st->clear.vs_layered);
+      st->clear.vs_layered = NULL;
+   }
+   if (st->clear.gs_layered) {
+      cso_delete_geometry_shader(st->cso_context, st->clear.gs_layered);
+      st->clear.gs_layered = NULL;
+   }
 }
 
 
 /**
  * Helper function to set the fragment shaders.
  */
-static INLINE void
+static inline void
 set_fragment_shader(struct st_context *st)
 {
-   if (!st->clear.fs)
-      st->clear.fs =
-         util_make_fragment_passthrough_shader(st->pipe, TGSI_SEMANTIC_GENERIC,
-                                               TGSI_INTERPOLATE_CONSTANT);
+   struct pipe_screen *pscreen = st->pipe->screen;
+   bool use_nir = PIPE_SHADER_IR_NIR ==
+      pscreen->get_shader_param(pscreen, PIPE_SHADER_VERTEX,
+                                PIPE_SHADER_CAP_PREFERRED_IR);
+
+   if (!st->clear.fs) {
+      if (use_nir) {
+         unsigned inputs[] = { VARYING_SLOT_VAR0 };
+         unsigned outputs[] = { FRAG_RESULT_COLOR };
+         unsigned interpolation[] = { INTERP_MODE_FLAT };
+         st->clear.fs = st_nir_make_passthrough_shader(st, "clear FS",
+                                                       MESA_SHADER_FRAGMENT,
+                                                       1, inputs, outputs,
+                                                       interpolation, 0);
+      } else {
+         st->clear.fs =
+            util_make_fragment_passthrough_shader(st->pipe,
+                                                  TGSI_SEMANTIC_GENERIC,
+                                                  TGSI_INTERPOLATE_CONSTANT,
+                                                  TRUE);
+      }
+   }
 
    cso_set_fragment_shader_handle(st->cso_context, st->clear.fs);
 }
 
 
+static void *
+make_nir_clear_vertex_shader(struct st_context *st, bool layered)
+{
+   const char *shader_name = layered ? "layered clear VS" : "clear VS";
+   unsigned inputs[] = {
+      VERT_ATTRIB_POS,
+      VERT_ATTRIB_GENERIC0,
+      SYSTEM_VALUE_INSTANCE_ID,
+   };
+   unsigned outputs[] = {
+      VARYING_SLOT_POS,
+      VARYING_SLOT_VAR0,
+      VARYING_SLOT_LAYER
+   };
+
+   return st_nir_make_passthrough_shader(st, shader_name, MESA_SHADER_VERTEX,
+                                         layered ? 3 : 2, inputs, outputs,
+                                         NULL, (1 << 2));
+}
+
+
 /**
  * Helper function to set the vertex shader.
  */
-static INLINE void
+static inline void
 set_vertex_shader(struct st_context *st)
 {
+   struct pipe_screen *pscreen = st->pipe->screen;
+   bool use_nir = PIPE_SHADER_IR_NIR ==
+      pscreen->get_shader_param(pscreen, PIPE_SHADER_VERTEX,
+                                PIPE_SHADER_CAP_PREFERRED_IR);
+
    /* vertex shader - still required to provide the linkage between
     * fragment shader input semantics and vertex_element/buffers.
     */
    if (!st->clear.vs)
    {
-      const uint semantic_names[] = { TGSI_SEMANTIC_POSITION,
-                                      TGSI_SEMANTIC_GENERIC };
-      const uint semantic_indexes[] = { 0, 0 };
-      st->clear.vs = util_make_vertex_passthrough_shader(st->pipe, 2,
-                                                         semantic_names,
-                                                         semantic_indexes);
+      if (use_nir) {
+         st->clear.vs = make_nir_clear_vertex_shader(st, false);
+      } else {
+         const enum tgsi_semantic semantic_names[] = {
+            TGSI_SEMANTIC_POSITION,
+            TGSI_SEMANTIC_GENERIC
+         };
+         const uint semantic_indexes[] = { 0, 0 };
+         st->clear.vs = util_make_vertex_passthrough_shader(st->pipe, 2,
+                                                            semantic_names,
+                                                            semantic_indexes,
+                                                            FALSE);
+      }
    }
 
    cso_set_vertex_shader_handle(st->cso_context, st->clear.vs);
+   cso_set_geometry_shader_handle(st->cso_context, NULL);
 }
 
 
-/**
- * Draw a screen-aligned quadrilateral.
- * Coords are clip coords with y=0=bottom.
- */
 static void
-draw_quad(struct st_context *st,
-          float x0, float y0, float x1, float y1, GLfloat z,
-          const union pipe_color_union *color)
+set_vertex_shader_layered(struct st_context *st)
 {
    struct pipe_context *pipe = st->pipe;
-   struct pipe_resource *vbuf = NULL;
-   GLuint i, offset;
-   float (*vertices)[2][4];  /**< vertex pos + color */
-
-   if (u_upload_alloc(st->uploader, 0, 4 * sizeof(vertices[0]),
-                      &offset, &vbuf, (void **) &vertices) != PIPE_OK) {
+   struct pipe_screen *pscreen = pipe->screen;
+   bool use_nir = PIPE_SHADER_IR_NIR ==
+      pscreen->get_shader_param(pscreen, PIPE_SHADER_VERTEX,
+                                PIPE_SHADER_CAP_PREFERRED_IR);
+
+   if (!pipe->screen->get_param(pipe->screen, PIPE_CAP_TGSI_INSTANCEID)) {
+      assert(!"Got layered clear, but VS instancing is unsupported");
+      set_vertex_shader(st);
       return;
    }
 
-   /* positions */
-   vertices[0][0][0] = x0;
-   vertices[0][0][1] = y0;
-
-   vertices[1][0][0] = x1;
-   vertices[1][0][1] = y0;
-
-   vertices[2][0][0] = x1;
-   vertices[2][0][1] = y1;
-
-   vertices[3][0][0] = x0;
-   vertices[3][0][1] = y1;
-
-   /* same for all verts: */
-   for (i = 0; i < 4; i++) {
-      vertices[i][0][2] = z;
-      vertices[i][0][3] = 1.0;
-      vertices[i][1][0] = color->f[0];
-      vertices[i][1][1] = color->f[1];
-      vertices[i][1][2] = color->f[2];
-      vertices[i][1][3] = color->f[3];
+   if (!st->clear.vs_layered) {
+      bool vs_layer =
+         pipe->screen->get_param(pipe->screen, PIPE_CAP_TGSI_VS_LAYER_VIEWPORT);
+      if (vs_layer) {
+         st->clear.vs_layered =
+            use_nir ? make_nir_clear_vertex_shader(st, true)
+                    : util_make_layered_clear_vertex_shader(pipe);
+      } else {
+         st->clear.vs_layered = util_make_layered_clear_helper_vertex_shader(pipe);
+         st->clear.gs_layered = util_make_layered_clear_geometry_shader(pipe);
+      }
    }
 
-   u_upload_unmap(st->uploader);
-
-   /* draw */
-   util_draw_vertex_buffer(pipe,
-                           st->cso_context,
-                           vbuf,
-                           cso_get_aux_vertex_buffer_slot(st->cso_context),
-                           offset,
-                           PIPE_PRIM_TRIANGLE_FAN,
-                           4,  /* verts */
-                           2); /* attribs/vert */
-
-   pipe_resource_reference(&vbuf, NULL);
+   cso_set_vertex_shader_handle(st->cso_context, st->clear.vs_layered);
+   cso_set_geometry_shader_handle(st->cso_context, st->clear.gs_layered);
 }
 
 
-
 /**
  * Do glClear by drawing a quadrilateral.
  * The vertices of the quad will be computed from the
  * ctx->DrawBuffer->_X/Ymin/max fields.
  */
 static void
-clear_with_quad(struct gl_context *ctx,
-                GLboolean color, GLboolean depth, GLboolean stencil)
+clear_with_quad(struct gl_context *ctx, unsigned clear_buffers)
 {
    struct st_context *st = st_context(ctx);
+   struct cso_context *cso = st->cso_context;
    const struct gl_framebuffer *fb = ctx->DrawBuffer;
    const GLfloat fb_width = (GLfloat) fb->Width;
    const GLfloat fb_height = (GLfloat) fb->Height;
+
+   _mesa_update_draw_buffer_bounds(ctx, ctx->DrawBuffer);
+
    const GLfloat x0 = (GLfloat) ctx->DrawBuffer->_Xmin / fb_width * 2.0f - 1.0f;
    const GLfloat x1 = (GLfloat) ctx->DrawBuffer->_Xmax / fb_width * 2.0f - 1.0f;
    const GLfloat y0 = (GLfloat) ctx->DrawBuffer->_Ymin / fb_height * 2.0f - 1.0f;
    const GLfloat y1 = (GLfloat) ctx->DrawBuffer->_Ymax / fb_height * 2.0f - 1.0f;
-   union pipe_color_union clearColor;
+   unsigned num_layers = st->state.fb_num_layers;
 
    /*
-   printf("%s %s%s%s %f,%f %f,%f\n", __FUNCTION__, 
+   printf("%s %s%s%s %f,%f %f,%f\n", __func__,
          color ? "color, " : "",
          depth ? "depth, " : "",
          stencil ? "stencil" : "",
@@ -214,24 +257,24 @@ clear_with_quad(struct gl_context *ctx,
          x1, y1);
    */
 
-   cso_save_blend(st->cso_context);
-   cso_save_stencil_ref(st->cso_context);
-   cso_save_depth_stencil_alpha(st->cso_context);
-   cso_save_rasterizer(st->cso_context);
-   cso_save_sample_mask(st->cso_context);
-   cso_save_viewport(st->cso_context);
-   cso_save_fragment_shader(st->cso_context);
-   cso_save_stream_outputs(st->cso_context);
-   cso_save_vertex_shader(st->cso_context);
-   cso_save_geometry_shader(st->cso_context);
-   cso_save_vertex_elements(st->cso_context);
-   cso_save_aux_vertex_buffer_slot(st->cso_context);
+   cso_save_state(cso, (CSO_BIT_BLEND |
+                        CSO_BIT_STENCIL_REF |
+                        CSO_BIT_DEPTH_STENCIL_ALPHA |
+                        CSO_BIT_RASTERIZER |
+                        CSO_BIT_SAMPLE_MASK |
+                        CSO_BIT_MIN_SAMPLES |
+                        CSO_BIT_VIEWPORT |
+                        CSO_BIT_STREAM_OUTPUTS |
+                        CSO_BIT_VERTEX_ELEMENTS |
+                        CSO_BIT_AUX_VERTEX_BUFFER_SLOT |
+                        (st->active_queries ? CSO_BIT_PAUSE_QUERIES : 0) |
+                        CSO_BITS_ALL_SHADERS));
 
    /* blend state: RGBA masking */
    {
       struct pipe_blend_state blend;
       memset(&blend, 0, sizeof(blend));
-      if (color) {
+      if (clear_buffers & PIPE_CLEAR_COLOR) {
          int num_buffers = ctx->Extensions.EXT_draw_buffers2 ?
                            ctx->DrawBuffer->_NumColorDrawBuffers : 1;
          int i;
@@ -239,33 +282,29 @@ clear_with_quad(struct gl_context *ctx,
          blend.independent_blend_enable = num_buffers > 1;
 
          for (i = 0; i < num_buffers; i++) {
-            if (ctx->Color.ColorMask[i][0])
-               blend.rt[i].colormask |= PIPE_MASK_R;
-            if (ctx->Color.ColorMask[i][1])
-               blend.rt[i].colormask |= PIPE_MASK_G;
-            if (ctx->Color.ColorMask[i][2])
-               blend.rt[i].colormask |= PIPE_MASK_B;
-            if (ctx->Color.ColorMask[i][3])
-               blend.rt[i].colormask |= PIPE_MASK_A;
+            if (!(clear_buffers & (PIPE_CLEAR_COLOR0 << i)))
+               continue;
+
+            blend.rt[i].colormask = GET_COLORMASK(ctx->Color.ColorMask, i);
          }
 
-         if (st->ctx->Color.DitherFlag)
+         if (ctx->Color.DitherFlag)
             blend.dither = 1;
       }
-      cso_set_blend(st->cso_context, &blend);
+      cso_set_blend(cso, &blend);
    }
 
    /* depth_stencil state: always pass/set to ref value */
    {
       struct pipe_depth_stencil_alpha_state depth_stencil;
       memset(&depth_stencil, 0, sizeof(depth_stencil));
-      if (depth) {
+      if (clear_buffers & PIPE_CLEAR_DEPTH) {
          depth_stencil.depth.enabled = 1;
          depth_stencil.depth.writemask = 1;
          depth_stencil.depth.func = PIPE_FUNC_ALWAYS;
       }
 
-      if (stencil) {
+      if (clear_buffers & PIPE_CLEAR_STENCIL) {
          struct pipe_stencil_ref stencil_ref;
          memset(&stencil_ref, 0, sizeof(stencil_ref));
          depth_stencil.stencil[0].enabled = 1;
@@ -276,96 +315,98 @@ clear_with_quad(struct gl_context *ctx,
          depth_stencil.stencil[0].valuemask = 0xff;
          depth_stencil.stencil[0].writemask = ctx->Stencil.WriteMask[0] & 0xff;
          stencil_ref.ref_value[0] = ctx->Stencil.Clear;
-         cso_set_stencil_ref(st->cso_context, &stencil_ref);
+         cso_set_stencil_ref(cso, &stencil_ref);
       }
 
-      cso_set_depth_stencil_alpha(st->cso_context, &depth_stencil);
+      cso_set_depth_stencil_alpha(cso, &depth_stencil);
    }
 
-   cso_set_vertex_elements(st->cso_context, 2, st->velems_util_draw);
-   cso_set_stream_outputs(st->cso_context, 0, NULL, 0);
-   cso_set_sample_mask(st->cso_context, ~0);
-   cso_set_rasterizer(st->cso_context, &st->clear.raster);
+   cso_set_vertex_elements(cso, 2, st->util_velems);
+   cso_set_stream_outputs(cso, 0, NULL, NULL);
+   cso_set_sample_mask(cso, ~0);
+   cso_set_min_samples(cso, 1);
+   st->clear.raster.multisample = st->state.fb_num_samples > 1;
+   cso_set_rasterizer(cso, &st->clear.raster);
 
    /* viewport state: viewport matching window dims */
-   {
-      const GLboolean invert = (st_fb_orientation(fb) == Y_0_TOP);
-      struct pipe_viewport_state vp;
-      vp.scale[0] = 0.5f * fb_width;
-      vp.scale[1] = fb_height * (invert ? -0.5f : 0.5f);
-      vp.scale[2] = 1.0f;
-      vp.scale[3] = 1.0f;
-      vp.translate[0] = 0.5f * fb_width;
-      vp.translate[1] = 0.5f * fb_height;
-      vp.translate[2] = 0.0f;
-      vp.translate[3] = 0.0f;
-      cso_set_viewport(st->cso_context, &vp);
-   }
+   cso_set_viewport_dims(st->cso_context, fb_width, fb_height,
+                         st_fb_orientation(fb) == Y_0_TOP);
 
    set_fragment_shader(st);
-   set_vertex_shader(st);
-   cso_set_geometry_shader_handle(st->cso_context, NULL);
-
-   if (ctx->DrawBuffer->_ColorDrawBuffers[0]) {
-      struct gl_renderbuffer *rb = ctx->DrawBuffer->_ColorDrawBuffers[0];
-      GLboolean is_integer = _mesa_is_enum_format_integer(rb->InternalFormat);
-
-      st_translate_color(&ctx->Color.ClearColor,
-                         &clearColor,
-                         ctx->DrawBuffer->_ColorDrawBuffers[0]->_BaseFormat,
-                         is_integer);
+   cso_set_tessctrl_shader_handle(cso, NULL);
+   cso_set_tesseval_shader_handle(cso, NULL);
+
+   if (num_layers > 1)
+      set_vertex_shader_layered(st);
+   else
+      set_vertex_shader(st);
+
+   /* draw quad matching scissor rect.
+    *
+    * Note: if we're only clearing depth/stencil we still setup vertices
+    * with color, but they'll be ignored.
+    *
+    * We can't translate the clear color to the colorbuffer format,
+    * because different colorbuffers may have different formats.
+    */
+   if (!st_draw_quad(st, x0, y0, x1, y1,
+                     ctx->Depth.Clear * 2.0f - 1.0f,
+                     0.0f, 0.0f, 0.0f, 0.0f,
+                     (const float *) &ctx->Color.ClearColor.f,
+                     num_layers)) {
+      _mesa_error(ctx, GL_OUT_OF_MEMORY, "glClear");
    }
 
-   /* draw quad matching scissor rect */
-   draw_quad(st, x0, y0, x1, y1, (GLfloat) ctx->Depth.Clear, &clearColor);
-
    /* Restore pipe state */
-   cso_restore_blend(st->cso_context);
-   cso_restore_stencil_ref(st->cso_context);
-   cso_restore_depth_stencil_alpha(st->cso_context);
-   cso_restore_rasterizer(st->cso_context);
-   cso_restore_sample_mask(st->cso_context);
-   cso_restore_viewport(st->cso_context);
-   cso_restore_fragment_shader(st->cso_context);
-   cso_restore_vertex_shader(st->cso_context);
-   cso_restore_geometry_shader(st->cso_context);
-   cso_restore_vertex_elements(st->cso_context);
-   cso_restore_aux_vertex_buffer_slot(st->cso_context);
-   cso_restore_stream_outputs(st->cso_context);
+   cso_restore_state(cso);
 }
 
 
 /**
  * Return if the scissor must be enabled during the clear.
  */
-static INLINE GLboolean
+static inline GLboolean
 is_scissor_enabled(struct gl_context *ctx, struct gl_renderbuffer *rb)
 {
-   return ctx->Scissor.Enabled &&
-          (ctx->Scissor.X > 0 ||
-           ctx->Scissor.Y > 0 ||
-           ctx->Scissor.Width < rb->Width ||
-           ctx->Scissor.Height < rb->Height);
+   const struct gl_scissor_rect *scissor = &ctx->Scissor.ScissorArray[0];
+
+   return (ctx->Scissor.EnableFlags & 1) &&
+          (scissor->X > 0 ||
+           scissor->Y > 0 ||
+           scissor->X + scissor->Width < (int)rb->Width ||
+           scissor->Y + scissor->Height < (int)rb->Height);
+}
+
+/**
+ * Return if window rectangles must be enabled during the clear.
+ */
+static inline bool
+is_window_rectangle_enabled(struct gl_context *ctx)
+{
+   if (ctx->DrawBuffer == ctx->WinSysDrawBuffer)
+      return false;
+   return ctx->Scissor.NumWindowRects > 0 ||
+      ctx->Scissor.WindowRectMode == GL_INCLUSIVE_EXT;
 }
 
 
 /**
- * Return if any of the color channels are masked.
+ * Return if all of the stencil bits are masked.
  */
-static INLINE GLboolean
-is_color_masked(struct gl_context *ctx, int i)
+static inline GLboolean
+is_stencil_disabled(struct gl_context *ctx, struct gl_renderbuffer *rb)
 {
-   return !ctx->Color.ColorMask[i][0] ||
-          !ctx->Color.ColorMask[i][1] ||
-          !ctx->Color.ColorMask[i][2] ||
-          !ctx->Color.ColorMask[i][3];
+   const GLuint stencilMax = 0xff;
+
+   assert(_mesa_get_format_bits(rb->Format, GL_STENCIL_BITS) > 0);
+   return (ctx->Stencil.WriteMask[0] & stencilMax) == 0;
 }
 
 
 /**
  * Return if any of the stencil bits are masked.
  */
-static INLINE GLboolean
+static inline GLboolean
 is_stencil_masked(struct gl_context *ctx, struct gl_renderbuffer *rb)
 {
    const GLuint stencilMax = 0xff;
@@ -390,14 +431,17 @@ st_Clear(struct gl_context *ctx, GLbitfield mask)
    GLbitfield clear_buffers = 0x0;
    GLuint i;
 
+   st_flush_bitmap_cache(st);
+   st_invalidate_readpix_cache(st);
+
    /* This makes sure the pipe has the latest scissor, etc values */
-   st_validate_state( st );
+   st_validate_state(st, ST_PIPELINE_CLEAR);
 
    if (mask & BUFFER_BITS_COLOR) {
       for (i = 0; i < ctx->DrawBuffer->_NumColorDrawBuffers; i++) {
-         GLuint b = ctx->DrawBuffer->_ColorDrawBufferIndexes[i];
+         gl_buffer_index b = ctx->DrawBuffer->_ColorDrawBufferIndexes[i];
 
-         if (mask & (1 << b)) {
+         if (b != BUFFER_NONE && mask & (1 << b)) {
             struct gl_renderbuffer *rb
                = ctx->DrawBuffer->Attachment[b].Renderbuffer;
             struct st_renderbuffer *strb = st_renderbuffer(rb);
@@ -406,11 +450,21 @@ st_Clear(struct gl_context *ctx, GLbitfield mask)
             if (!strb || !strb->surface)
                continue;
 
+            unsigned colormask =
+               GET_COLORMASK(ctx->Color.ColorMask, colormask_index);
+
+            if (!colormask)
+               continue;
+
+            unsigned surf_colormask =
+               util_format_colormask(util_format_description(strb->surface->format));
+
             if (is_scissor_enabled(ctx, rb) ||
-                is_color_masked(ctx, colormask_index))
-               quad_buffers |= PIPE_CLEAR_COLOR;
+                is_window_rectangle_enabled(ctx) ||
+                ((colormask & surf_colormask) != surf_colormask))
+               quad_buffers |= PIPE_CLEAR_COLOR0 << i;
             else
-               clear_buffers |= PIPE_CLEAR_COLOR;
+               clear_buffers |= PIPE_CLEAR_COLOR0 << i;
          }
       }
    }
@@ -418,8 +472,9 @@ st_Clear(struct gl_context *ctx, GLbitfield mask)
    if (mask & BUFFER_BIT_DEPTH) {
       struct st_renderbuffer *strb = st_renderbuffer(depthRb);
 
-      if (strb->surface) {
-         if (is_scissor_enabled(ctx, depthRb))
+      if (strb->surface && ctx->Depth.Mask) {
+         if (is_scissor_enabled(ctx, depthRb) ||
+             is_window_rectangle_enabled(ctx))
             quad_buffers |= PIPE_CLEAR_DEPTH;
          else
             clear_buffers |= PIPE_CLEAR_DEPTH;
@@ -428,8 +483,9 @@ st_Clear(struct gl_context *ctx, GLbitfield mask)
    if (mask & BUFFER_BIT_STENCIL) {
       struct st_renderbuffer *strb = st_renderbuffer(stencilRb);
 
-      if (strb->surface) {
+      if (strb->surface && !is_stencil_disabled(ctx, stencilRb)) {
          if (is_scissor_enabled(ctx, stencilRb) ||
+             is_window_rectangle_enabled(ctx) ||
              is_stencil_masked(ctx, stencilRb))
             quad_buffers |= PIPE_CLEAR_STENCIL;
          else
@@ -437,32 +493,30 @@ st_Clear(struct gl_context *ctx, GLbitfield mask)
       }
    }
 
-   /*
-    * If we're going to use clear_with_quad() for any reason, use it for
-    * everything possible.
+   /* Always clear depth and stencil together.
+    * This can only happen when the stencil writemask is not a full mask.
     */
-   if (quad_buffers) {
-      quad_buffers |= clear_buffers;
-      clear_with_quad(ctx,
-                      quad_buffers & PIPE_CLEAR_COLOR,
-                      quad_buffers & PIPE_CLEAR_DEPTH,
-                      quad_buffers & PIPE_CLEAR_STENCIL);
-   } else if (clear_buffers) {
-      union pipe_color_union clearColor;
-
-      if (ctx->DrawBuffer->_ColorDrawBuffers[0]) {
-         struct gl_renderbuffer *rb = ctx->DrawBuffer->_ColorDrawBuffers[0];
-         GLboolean is_integer = _mesa_is_enum_format_integer(rb->InternalFormat);
-
-         st_translate_color(&ctx->Color.ClearColor,
-                            &clearColor,
-                           ctx->DrawBuffer->_ColorDrawBuffers[0]->_BaseFormat,
-                           is_integer);
-      }
+   if (quad_buffers & PIPE_CLEAR_DEPTHSTENCIL &&
+       clear_buffers & PIPE_CLEAR_DEPTHSTENCIL) {
+      quad_buffers |= clear_buffers & PIPE_CLEAR_DEPTHSTENCIL;
+      clear_buffers &= ~PIPE_CLEAR_DEPTHSTENCIL;
+   }
 
-      st->pipe->clear(st->pipe, clear_buffers, &clearColor,
+   /* Only use quad-based clearing for the renderbuffers which cannot
+    * use pipe->clear. We want to always use pipe->clear for the other
+    * renderbuffers, because it's likely to be faster.
+    */
+   if (clear_buffers) {
+      /* We can't translate the clear color to the colorbuffer format,
+       * because different colorbuffers may have different formats.
+       */
+      st->pipe->clear(st->pipe, clear_buffers,
+                      (union pipe_color_union*)&ctx->Color.ClearColor,
                       ctx->Depth.Clear, ctx->Stencil.Clear);
    }
+   if (quad_buffers) {
+      clear_with_quad(ctx, quad_buffers);
+   }
    if (mask & BUFFER_BIT_ACCUM)
       _mesa_clear_accum_buffer(ctx);
 }