intel: Add a batch flush between front-buffer downsample and X protocol.

[mesa.git] / src / mesa / drivers / dri / i965 / brw_shader.cpp

diff --git a/src/mesa/drivers/dri/i965/brw_shader.cpp b/src/mesa/drivers/dri/i965/brw_shader.cpp
index 03e432970232ba555c94c66ecf7acbbb3f1405d6..e7dbdbe89207f4322967c17b7e3f73fe92aeaf2d 100644 (file)
--- a/src/mesa/drivers/dri/i965/brw_shader.cpp
+++ b/src/mesa/drivers/dri/i965/brw_shader.cpp
@@ -24,11 +24,12 @@
 extern "C" {
 #include "main/macros.h"
 #include "brw_context.h"
-#include "brw_vs.h"
 }
+#include "brw_vs.h"
 #include "brw_fs.h"
 #include "glsl/ir_optimization.h"
-#include "glsl/ir_print_visitor.h"
+#include "glsl/glsl_parser_extras.h"
+#include "main/shaderapi.h"
 
 struct gl_shader *
 brw_new_shader(struct gl_context *ctx, GLuint name, GLuint type)
@@ -89,7 +90,7 @@ brw_lower_packing_builtins(struct brw_context *brw,
            | LOWER_PACK_UNORM_4x8
            | LOWER_UNPACK_UNORM_4x8;
 
-   if (brw->intel.gen >= 7) {
+   if (brw->gen >= 7) {
       /* Gen7 introduced the f32to16 and f16to32 instructions, which can be
        * used to execute packHalf2x16 and unpackHalf2x16. For AOS code, no
        * lowering is needed. For SOA code, the Half2x16 ops must be
@@ -111,33 +112,23 @@ GLboolean
 brw_link_shader(struct gl_context *ctx, struct gl_shader_program *shProg)
 {
    struct brw_context *brw = brw_context(ctx);
-   struct intel_context *intel = &brw->intel;
    unsigned int stage;
-   static const char *target_strings[]
-      = { "vertex", "fragment", "geometry" };
 
    for (stage = 0; stage < ARRAY_SIZE(shProg->_LinkedShaders); stage++) {
       struct brw_shader *shader =
         (struct brw_shader *)shProg->_LinkedShaders[stage];
-      static const GLenum targets[] = {
-        GL_VERTEX_PROGRAM_ARB,
-        GL_FRAGMENT_PROGRAM_ARB,
-        GL_GEOMETRY_PROGRAM_NV
-      };
 
       if (!shader)
         continue;
 
       struct gl_program *prog =
-        ctx->Driver.NewProgram(ctx, targets[stage], shader->base.Name);
+        ctx->Driver.NewProgram(ctx, _mesa_program_index_to_target(stage),
+                                shader->base.Name);
       if (!prog)
        return false;
       prog->Parameters = _mesa_new_parameter_list();
 
-      if (stage == 0) {
-        struct gl_vertex_program *vp = (struct gl_vertex_program *) prog;
-        vp->UsesClipDistance = shProg->Vert.UsesClipDistance;
-      }
+      _mesa_copy_linked_program_data((gl_shader_type) stage, shProg, prog);
 
       void *mem_ctx = ralloc_context(NULL);
       bool progress;
@@ -152,10 +143,10 @@ brw_link_shader(struct gl_context *ctx, struct gl_shader_program *shProg)
        */
       brw_lower_packing_builtins(brw, (gl_shader_type) stage, shader->ir);
       do_mat_op_to_vec(shader->ir);
-      const int bitfield_insert = intel->gen >= 7
+      const int bitfield_insert = brw->gen >= 7
                                   ? BITFIELD_INSERT_TO_BFM_BFI
                                   : 0;
-      const int lrp_to_arith = intel->gen < 6 ? LRP_TO_ARITH : 0;
+      const int lrp_to_arith = brw->gen < 6 ? LRP_TO_ARITH : 0;
       lower_instructions(shader->ir,
                         MOD_TO_FRACT |
                         DIV_TO_MUL_RCP |
@@ -168,11 +159,11 @@ brw_link_shader(struct gl_context *ctx, struct gl_shader_program *shProg)
       /* Pre-gen6 HW can only nest if-statements 16 deep.  Beyond this,
        * if-statements need to be flattened.
        */
-      if (intel->gen < 6)
+      if (brw->gen < 6)
         lower_if_to_cond_assign(shader->ir, 16);
 
       do_lower_texture_projection(shader->ir);
-      brw_lower_texture_gradients(intel, shader->ir);
+      brw_lower_texture_gradients(brw, shader->ir);
       do_vec_index_to_cond_assign(shader->ir);
       lower_vector_insert(shader->ir, true);
       brw_do_cubemap_normalize(shader->ir);
@@ -188,7 +179,7 @@ brw_link_shader(struct gl_context *ctx, struct gl_shader_program *shProg)
          lower_variable_index_to_cond_assign(shader->ir,
                                              input, output, temp, uniform);
 
-      if (unlikely((intel->perf_debug) && lowered_variable_indexing)) {
+      if (unlikely(brw->perf_debug && lowered_variable_indexing)) {
          perf_debug("Unsupported form of variable indexing in FS; falling "
                     "back to very inefficient code generation\n");
       }
@@ -243,8 +234,7 @@ brw_link_shader(struct gl_context *ctx, struct gl_shader_program *shProg)
       reparent_ir(shader->ir, shader->ir);
       ralloc_free(mem_ctx);
 
-      do_set_program_inouts(shader->ir, prog,
-                           shader->base.Type == GL_FRAGMENT_SHADER);
+      do_set_program_inouts(shader->ir, prog, shader->base.Type);
 
       prog->SamplersUsed = shader->base.active_samplers;
       _mesa_update_shader_textures_used(shProg, prog);
@@ -263,8 +253,8 @@ brw_link_shader(struct gl_context *ctx, struct gl_shader_program *shProg)
 
       if (ctx->Shader.Flags & GLSL_DUMP) {
          printf("\n");
-         printf("GLSL IR for linked %s program %d:\n", target_strings[stage],
-                shProg->Name);
+         printf("GLSL IR for linked %s program %d:\n",
+                _mesa_glsl_shader_target_name(shader->base.Type), shProg->Name);
          _mesa_print_ir(shader->base.ir, NULL);
          printf("\n");
       }
@@ -277,7 +267,7 @@ brw_link_shader(struct gl_context *ctx, struct gl_shader_program *shProg)
             continue;
 
          printf("GLSL %s shader %d source for linked program %d:\n",
-                target_strings[_mesa_shader_type_to_index(sh->Type)],
+                _mesa_glsl_shader_target_name(sh->Type),
                 i,
                 shProg->Name);
          printf("%s", sh->Source);
@@ -495,7 +485,7 @@ brw_instruction_name(enum opcode op)
       return "placeholder_halt";
 
    case VS_OPCODE_URB_WRITE:
-      return "urb_write";
+      return "vs_urb_write";
    case VS_OPCODE_SCRATCH_READ:
       return "scratch_read";
    case VS_OPCODE_SCRATCH_WRITE:
@@ -504,6 +494,19 @@ brw_instruction_name(enum opcode op)
       return "pull_constant_load";
    case VS_OPCODE_PULL_CONSTANT_LOAD_GEN7:
       return "pull_constant_load_gen7";
+   case VS_OPCODE_UNPACK_FLAGS_SIMD4X2:
+      return "unpack_flags_simd4x2";
+
+   case GS_OPCODE_URB_WRITE:
+      return "gs_urb_write";
+   case GS_OPCODE_THREAD_END:
+      return "gs_thread_end";
+   case GS_OPCODE_SET_WRITE_OFFSET:
+      return "set_write_offset";
+   case GS_OPCODE_SET_VERTEX_COUNT:
+      return "set_vertex_count";
+   case GS_OPCODE_SET_DWORD_2_IMMED:
+      return "set_dword_2_immed";
 
    default:
       /* Yes, this leaks.  It's in debug code, it should never occur, and if