[Mesa-dev] [PATCH 1/3] i965: Move SOL PSIZ hacks from draw time to link time.

[Kenneth Graunke]

We can just update the gl_transform_feedback_info fields at link time
to make the VUE header fields have the right location and component.
Then we don't need to handle them specially at draw time, which is
expensive.
---
 src/intel/compiler/gen6_gs_visitor.cpp        | 13 +---------
 src/mesa/drivers/dri/i965/brw_link.cpp        | 36 +++++++++++++++++++++++++++
 src/mesa/drivers/dri/i965/genX_state_upload.c | 26 +++----------------
 3 files changed, 41 insertions(+), 34 deletions(-)

Here are a few cleanups to the SO_DECL handling I wrote a few months ago
while reading through the state upload code.  Really, we should move the
SO_DECL_LIST processing to link-time...there's no current API state in
there.  Ideally we could just memcpy it all to the batch at draw time...

diff --git a/src/intel/compiler/gen6_gs_visitor.cpp b/src/intel/compiler/gen6_gs_visitor.cpp
index fe9f834f0ce..66c69fb058f 100644
--- a/src/intel/compiler/gen6_gs_visitor.cpp
+++ b/src/intel/compiler/gen6_gs_visitor.cpp
@@ -687,18 +687,7 @@ gen6_gs_visitor::xfb_program(unsigned vertex, unsigned num_verts)
          emit(MOV(dst_reg(this->vertex_output_offset), brw_imm_d(offset)));
          memcpy(data.reladdr, &this->vertex_output_offset, sizeof(src_reg));
          data.type = output_reg[varying][0].type;
-
-         /* PSIZ, LAYER and VIEWPORT are packed in different channels of the
-          * same slot, so make sure we write the appropriate channel
-          */
-         if (varying == VARYING_SLOT_PSIZ)
-            data.swizzle = BRW_SWIZZLE_WWWW;
-         else if (varying == VARYING_SLOT_LAYER)
-            data.swizzle = BRW_SWIZZLE_YYYY;
-         else if (varying == VARYING_SLOT_VIEWPORT)
-            data.swizzle = BRW_SWIZZLE_ZZZZ;
-         else
-            data.swizzle = gs_prog_data->transform_feedback_swizzles[binding];
+         data.swizzle = gs_prog_data->transform_feedback_swizzles[binding];
 
          /* Write data */
          inst = emit(GS_OPCODE_SVB_WRITE, mrf_reg, data, sol_temp);
diff --git a/src/mesa/drivers/dri/i965/brw_link.cpp b/src/mesa/drivers/dri/i965/brw_link.cpp
index 57aaf6b9dc0..e9158c596c5 100644
--- a/src/mesa/drivers/dri/i965/brw_link.cpp
+++ b/src/mesa/drivers/dri/i965/brw_link.cpp
@@ -28,6 +28,7 @@
 #include "compiler/glsl/ir_optimization.h"
 #include "compiler/glsl/program.h"
 #include "program/program.h"
+#include "main/mtypes.h"
 #include "main/shaderapi.h"
 #include "main/shaderobj.h"
 #include "main/uniforms.h"
@@ -176,6 +177,39 @@ unify_interfaces(struct shader_info **infos)
    }
 }
 
+static void
+update_xfb_info(struct gl_transform_feedback_info *xfb_info)
+{
+   if (!xfb_info)
+      return;
+
+   for (unsigned i = 0; i < xfb_info->NumOutputs; i++) {
+      struct gl_transform_feedback_output *output = &xfb_info->Outputs[i];
+
+      /* The VUE header contains three scalar fields packed together:
+       * - gl_PointSize is stored in VARYING_SLOT_PSIZ.w
+       * - gl_Layer is stored in VARYING_SLOT_PSIZ.y
+       * - gl_ViewportIndex is stored in VARYING_SLOT_PSIZ.z
+       */
+      switch (output->OutputRegister) {
+      case VARYING_SLOT_LAYER:
+         assert(output->NumComponents == 1);
+         output->OutputRegister = VARYING_SLOT_PSIZ;
+         output->ComponentOffset = 1;
+         break;
+      case VARYING_SLOT_VIEWPORT:
+         assert(output->NumComponents == 1);
+         output->OutputRegister = VARYING_SLOT_PSIZ;
+         output->ComponentOffset = 2;
+         break;
+      case VARYING_SLOT_PSIZ:
+         assert(output->NumComponents == 1);
+         output->ComponentOffset = 3;
+         break;
+      }
+   }
+}
+
 extern "C" GLboolean
 brw_link_shader(struct gl_context *ctx, struct gl_shader_program *shProg)
 {
@@ -199,6 +233,8 @@ brw_link_shader(struct gl_context *ctx, struct gl_shader_program *shProg)
       prog->ShadowSamplers = shader->shadow_samplers;
       _mesa_update_shader_textures_used(shProg, prog);
 
+      update_xfb_info(prog->sh.LinkedTransformFeedback);
+
       bool debug_enabled =
          (INTEL_DEBUG & intel_debug_flag_for_shader_stage(shader->Stage));
 
diff --git a/src/mesa/drivers/dri/i965/genX_state_upload.c b/src/mesa/drivers/dri/i965/genX_state_upload.c
index 76d2ea887b1..a2ed2e72710 100644
--- a/src/mesa/drivers/dri/i965/genX_state_upload.c
+++ b/src/mesa/drivers/dri/i965/genX_state_upload.c
@@ -3087,32 +3087,14 @@ genX(upload_3dstate_so_decl_list)(struct brw_context *brw,
       unsigned decl_buffer_slot = buffer;
       assert(stream_id < MAX_VERTEX_STREAMS);
 
-      /* gl_PointSize is stored in VARYING_SLOT_PSIZ.w
-       * gl_Layer is stored in VARYING_SLOT_PSIZ.y
-       * gl_ViewportIndex is stored in VARYING_SLOT_PSIZ.z
-       */
-      if (varying == VARYING_SLOT_PSIZ) {
-         assert(components == 1);
-         component_mask <<= 3;
-      } else if (varying == VARYING_SLOT_LAYER) {
-         assert(components == 1);
-         component_mask <<= 1;
-      } else if (varying == VARYING_SLOT_VIEWPORT) {
-         assert(components == 1);
-         component_mask <<= 2;
-      } else {
-         component_mask <<= linked_xfb_info->Outputs[i].ComponentOffset;
-      }
+      component_mask <<= linked_xfb_info->Outputs[i].ComponentOffset;
 
       buffer_mask[stream_id] |= 1 << buffer;
 
+      assert(vue_map->varying_to_slot[varying] >= 0);
+
       decl.OutputBufferSlot = decl_buffer_slot;
-      if (varying == VARYING_SLOT_LAYER || varying == VARYING_SLOT_VIEWPORT) {
-         decl.RegisterIndex = vue_map->varying_to_slot[VARYING_SLOT_PSIZ];
-      } else {
-         assert(vue_map->varying_to_slot[varying] >= 0);
-         decl.RegisterIndex = vue_map->varying_to_slot[varying];
-      }
+      decl.RegisterIndex = vue_map->varying_to_slot[varying];
       decl.ComponentMask = component_mask;
 
       /* Mesa doesn't store entries for gl_SkipComponents in the Outputs[]
-- 
2.13.0