i965: Make the fragment shader pull constants index by dwords, not vec4s.

We want to load vec4s, since loading a vec4 instead of a dword is
basically no increased latency.  But for variable indexed access, the
previous requirement of aligned vec4s for a sampler LD was hard to
implement.

Note that this change only affects those messages that use the surface
format, like sampler LDs, but not to the untyped data cache loads we've
used in other cases.

No significant performance difference on my GLSL demo with uniforms forced
to take the varying pull constants path (n=4).

NOTE: This is a candidate for the 9.1 branch.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

src/mesa/drivers/dri/i965/brw_fs.cpp

@@ -2478,10 +2478,13 @@ fs_visitor::lower_uniform_pull_constant_loads()
          continue;
 
       if (intel->gen >= 7) {
+         /* The offset arg before was a vec4-aligned byte offset.  We need to
+          * turn it into a dword offset.
+          */
          fs_reg const_offset_reg = inst->src[1];
          assert(const_offset_reg.file == IMM &&
                 const_offset_reg.type == BRW_REGISTER_TYPE_UD);
-         const_offset_reg.imm.u /= 16;
+         const_offset_reg.imm.u /= 4;
          fs_reg payload = fs_reg(this, glsl_type::uint_type);
 
          /* This is actually going to be a MOV, but since only the first dword

src/mesa/drivers/dri/i965/brw_state.h

@@ -187,11 +187,6 @@ void *brw_state_batch(struct brw_context *brw,
 void gen4_init_vtable_surface_functions(struct brw_context *brw);
 uint32_t brw_get_surface_tiling_bits(uint32_t tiling);
 uint32_t brw_get_surface_num_multisamples(unsigned num_samples);
-void brw_create_constant_surface(struct brw_context *brw,
-				 drm_intel_bo *bo,
-				 uint32_t offset,
-				 int width,
-				 uint32_t *out_offset);
 
 uint32_t brw_format_for_mesa_format(gl_format mesa_format);
 

src/mesa/drivers/dri/i965/brw_vs_surface_state.c

@@ -91,7 +91,7 @@ brw_upload_vs_pull_constants(struct brw_context *brw)
 
    const int surf = SURF_INDEX_VERT_CONST_BUFFER;
    intel->vtbl.create_constant_surface(brw, brw->vs.const_bo, 0, size,
-				       &brw->vs.surf_offset[surf]);
+				       &brw->vs.surf_offset[surf], false);
 
    brw->state.dirty.brw |= BRW_NEW_VS_CONSTBUF;
 }

src/mesa/drivers/dri/i965/brw_wm_surface_state.c

@@ -913,15 +913,16 @@ brw_update_texture_surface(struct gl_context *ctx,
  * Create the constant buffer surface.  Vertex/fragment shader constants will be
  * read from this buffer with Data Port Read instructions/messages.
  */
-void
+static void
 brw_create_constant_surface(struct brw_context *brw,
 			    drm_intel_bo *bo,
 			    uint32_t offset,
 			    uint32_t size,
-			    uint32_t *out_offset)
+			    uint32_t *out_offset,
+                            bool dword_pitch)
 {
    struct intel_context *intel = &brw->intel;
-   uint32_t stride = 16;
+   uint32_t stride = dword_pitch ? 4 : 16;
    uint32_t elements = ALIGN(size, stride) / stride;
    const GLint w = elements - 1;
    uint32_t *surf;
@@ -1090,7 +1091,8 @@ brw_upload_wm_pull_constants(struct brw_context *brw)
    drm_intel_gem_bo_unmap_gtt(brw->wm.const_bo);
 
    intel->vtbl.create_constant_surface(brw, brw->wm.const_bo, 0, size,
-				       &brw->wm.surf_offset[surf_index]);
+				       &brw->wm.surf_offset[surf_index],
+                                       true);
 
    brw->state.dirty.brw |= BRW_NEW_SURFACES;
 }
@@ -1443,7 +1445,8 @@ brw_upload_ubo_surfaces(struct brw_context *brw,
        */
       intel->vtbl.create_constant_surface(brw, bo, binding->Offset,
 					  bo->size - binding->Offset,
-					  &surf_offsets[i]);
+					  &surf_offsets[i],
+                                          shader->Type == GL_FRAGMENT_SHADER);
    }
 
    if (shader->NumUniformBlocks)

src/mesa/drivers/dri/i965/gen7_wm_surface_state.c

@@ -384,10 +384,11 @@ gen7_create_constant_surface(struct brw_context *brw,
 			     drm_intel_bo *bo,
 			     uint32_t offset,
 			     uint32_t size,
-			     uint32_t *out_offset)
+			     uint32_t *out_offset,
+                             bool dword_pitch)
 {
    struct intel_context *intel = &brw->intel;
-   uint32_t stride = 16;
+   uint32_t stride = dword_pitch ? 4 : 16;
    uint32_t elements = ALIGN(size, stride) / stride;
    const GLint w = elements - 1;
 

src/mesa/drivers/dri/intel/intel_context.h

@@ -203,13 +203,14 @@ struct intel_context
 				      drm_intel_bo *bo,
 				      uint32_t offset,
 				      uint32_t size,
-				      uint32_t *out_offset);
+				      uint32_t *out_offset,
+                                      bool dword_pitch);
       /** \} */
    } vtbl;
 
    GLbitfield Fallback;  /**< mask of INTEL_FALLBACK_x bits */
    GLuint NewGLState;
-
+ 
    dri_bufmgr *bufmgr;
    unsigned int maxBatchSize;