i965: use new subroutine index uploader.
[mesa.git] / src / mesa / drivers / dri / i965 / brw_vs_surface_state.c
index 216ff4734bc75b143f0346f53d8f9daf92d4ffd1..387a54bdb0b4fc346ca0a0887e7fa0b71377d361 100644 (file)
@@ -1,6 +1,6 @@
 /*
  Copyright (C) Intel Corp.  2006.  All Rights Reserved.
- Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ Intel funded Tungsten Graphics to
  develop this 3D driver.
 
  Permission is hereby granted, free of charge, to any person obtaining
  **********************************************************************/
  /*
   * Authors:
-  *   Keith Whitwell <keith@tungstengraphics.com>
+  *   Keith Whitwell <keithw@vmware.com>
   */
 
 #include "main/mtypes.h"
 #include "program/prog_parameter.h"
+#include "main/shaderapi.h"
 
 #include "brw_context.h"
 #include "brw_state.h"
+#include "intel_buffer_objects.h"
 
-
+/**
+ * Creates a temporary BO containing the pull constant data for the shader
+ * stage, and the SURFACE_STATE struct that points at it.
+ *
+ * Pull constants are GLSL uniforms (and other constant data) beyond what we
+ * could fit as push constants, or that have variable-index array access
+ * (which is easiest to support using pull constants, and avoids filling
+ * register space with mostly-unused data).
+ *
+ * Compare this path to brw_curbe.c for gen4/5 push constants, and
+ * gen6_vs_state.c for gen6+ push constants.
+ */
 void
-brw_upload_vec4_pull_constants(struct brw_context *brw,
-                               GLbitfield brw_new_constbuf,
-                               const struct gl_program *prog,
-                               struct brw_stage_state *stage_state,
-                               const struct brw_vec4_prog_data *prog_data)
+brw_upload_pull_constants(struct brw_context *brw,
+                          GLbitfield64 brw_new_constbuf,
+                          const struct gl_program *prog,
+                          struct brw_stage_state *stage_state,
+                          const struct brw_stage_prog_data *prog_data)
 {
-   int i;
-
-   /* Updates the ParamaterValues[i] pointers for all parameters of the
-    * basic type of PROGRAM_STATE_VAR.
-    */
-   _mesa_load_state_parameters(&brw->ctx, prog->Parameters);
+   unsigned i;
+   uint32_t surf_index = prog_data->binding_table.pull_constants_start;
 
    if (!prog_data->nr_pull_params) {
-      if (stage_state->const_bo) {
-        drm_intel_bo_unreference(stage_state->const_bo);
-        stage_state->const_bo = NULL;
-        stage_state->surf_offset[SURF_INDEX_VEC4_CONST_BUFFER] = 0;
-        brw->state.dirty.brw |= brw_new_constbuf;
+      if (stage_state->surf_offset[surf_index]) {
+        stage_state->surf_offset[surf_index] = 0;
+        brw->ctx.NewDriverState |= brw_new_constbuf;
       }
       return;
    }
 
-   /* _NEW_PROGRAM_CONSTANTS */
-   drm_intel_bo_unreference(stage_state->const_bo);
+   /* Updates the ParamaterValues[i] pointers for all parameters of the
+    * basic type of PROGRAM_STATE_VAR.
+    */
+   _mesa_load_state_parameters(&brw->ctx, prog->Parameters);
+
+   /* BRW_NEW_*_PROG_DATA | _NEW_PROGRAM_CONSTANTS */
    uint32_t size = prog_data->nr_pull_params * 4;
-   stage_state->const_bo = drm_intel_bo_alloc(brw->bufmgr, "vec4_const_buffer",
-                                           size, 64);
+   drm_intel_bo *const_bo = NULL;
+   uint32_t const_offset;
+   gl_constant_value *constants = intel_upload_space(brw, size, 64,
+                                                     &const_bo, &const_offset);
 
-   drm_intel_gem_bo_map_gtt(stage_state->const_bo);
+   STATIC_ASSERT(sizeof(gl_constant_value) == sizeof(float));
 
    for (i = 0; i < prog_data->nr_pull_params; i++) {
-      memcpy(stage_state->const_bo->virtual + i * 4,
-            prog_data->pull_param[i],
-            4);
+      constants[i] = *prog_data->pull_param[i];
    }
 
    if (0) {
       for (i = 0; i < ALIGN(prog_data->nr_pull_params, 4) / 4; i++) {
-        float *row = (float *)stage_state->const_bo->virtual + i * 4;
-        printf("const surface %3d: %4.3f %4.3f %4.3f %4.3f\n",
-               i, row[0], row[1], row[2], row[3]);
+        const gl_constant_value *row = &constants[i * 4];
+        fprintf(stderr, "const surface %3d: %4.3f %4.3f %4.3f %4.3f\n",
+                 i, row[0].f, row[1].f, row[2].f, row[3].f);
       }
    }
 
-   drm_intel_gem_bo_unmap_gtt(stage_state->const_bo);
+   brw_create_constant_surface(brw, const_bo, const_offset, size,
+                               &stage_state->surf_offset[surf_index]);
+   drm_intel_bo_unreference(const_bo);
 
-   const int surf = SURF_INDEX_VEC4_CONST_BUFFER;
-   brw->vtbl.create_constant_surface(brw, stage_state->const_bo, 0, size,
-                                     &stage_state->surf_offset[surf], false);
-
-   brw->state.dirty.brw |= brw_new_constbuf;
+   brw->ctx.NewDriverState |= brw_new_constbuf;
 }
 
 
@@ -107,19 +116,22 @@ brw_upload_vs_pull_constants(struct brw_context *brw)
    struct brw_vertex_program *vp =
       (struct brw_vertex_program *) brw->vertex_program;
 
-   /* CACHE_NEW_VS_PROG */
-   const struct brw_vec4_prog_data *prog_data = &brw->vs.prog_data->base;
+   /* BRW_NEW_VS_PROG_DATA */
+   const struct brw_stage_prog_data *prog_data = &brw->vs.prog_data->base.base;
 
+   _mesa_shader_write_subroutine_indices(&brw->ctx, MESA_SHADER_VERTEX);
    /* _NEW_PROGRAM_CONSTANTS */
-   brw_upload_vec4_pull_constants(brw, BRW_NEW_VS_CONSTBUF, &vp->program.Base,
-                                  stage_state, prog_data);
+   brw_upload_pull_constants(brw, BRW_NEW_VS_CONSTBUF, &vp->program.Base,
+                             stage_state, prog_data);
 }
 
 const struct brw_tracked_state brw_vs_pull_constants = {
    .dirty = {
-      .mesa = (_NEW_PROGRAM_CONSTANTS),
-      .brw = (BRW_NEW_BATCH | BRW_NEW_VERTEX_PROGRAM),
-      .cache = CACHE_NEW_VS_PROG,
+      .mesa = _NEW_PROGRAM_CONSTANTS,
+      .brw = BRW_NEW_BATCH |
+             BRW_NEW_BLORP |
+             BRW_NEW_VERTEX_PROGRAM |
+             BRW_NEW_VS_PROG_DATA,
    },
    .emit = brw_upload_vs_pull_constants,
 };
@@ -127,85 +139,79 @@ const struct brw_tracked_state brw_vs_pull_constants = {
 static void
 brw_upload_vs_ubo_surfaces(struct brw_context *brw)
 {
-   struct brw_stage_state *stage_state = &brw->vs.base;
-
    struct gl_context *ctx = &brw->ctx;
    /* _NEW_PROGRAM */
-   struct gl_shader_program *prog = ctx->Shader.CurrentVertexProgram;
+   struct gl_shader_program *prog =
+      ctx->_Shader->CurrentProgram[MESA_SHADER_VERTEX];
 
    if (!prog)
       return;
 
+   /* BRW_NEW_VS_PROG_DATA */
    brw_upload_ubo_surfaces(brw, prog->_LinkedShaders[MESA_SHADER_VERTEX],
-                          &stage_state->surf_offset[SURF_INDEX_VEC4_UBO(0)]);
+                           &brw->vs.base, &brw->vs.prog_data->base.base);
 }
 
 const struct brw_tracked_state brw_vs_ubo_surfaces = {
    .dirty = {
       .mesa = _NEW_PROGRAM,
-      .brw = BRW_NEW_BATCH | BRW_NEW_UNIFORM_BUFFER,
-      .cache = 0,
+      .brw = BRW_NEW_BATCH |
+             BRW_NEW_BLORP |
+             BRW_NEW_UNIFORM_BUFFER |
+             BRW_NEW_VS_PROG_DATA,
    },
    .emit = brw_upload_vs_ubo_surfaces,
 };
 
-
-void
-brw_vec4_upload_binding_table(struct brw_context *brw,
-                              GLbitfield brw_new_binding_table,
-                              struct brw_stage_state *stage_state,
-                              const struct brw_vec4_prog_data *prog_data)
+static void
+brw_upload_vs_abo_surfaces(struct brw_context *brw)
 {
-   if (INTEL_DEBUG & DEBUG_SHADER_TIME) {
-      gen7_create_shader_time_surface(brw, &stage_state->surf_offset[SURF_INDEX_VEC4_SHADER_TIME]);
-   }
+   struct gl_context *ctx = &brw->ctx;
+   /* _NEW_PROGRAM */
+   struct gl_shader_program *prog =
+      ctx->_Shader->CurrentProgram[MESA_SHADER_VERTEX];
 
-   /* If there are no surfaces, skip making the binding table altogether. */
-   const unsigned entries = prog_data->binding_table_size;
-   if (entries == 0) {
-      if (stage_state->bind_bo_offset != 0) {
-        brw->state.dirty.brw |= brw_new_binding_table;
-        stage_state->bind_bo_offset = 0;
-      }
-      return;
+   if (prog) {
+      /* BRW_NEW_VS_PROG_DATA */
+      brw_upload_abo_surfaces(brw, prog->_LinkedShaders[MESA_SHADER_VERTEX],
+                              &brw->vs.base, &brw->vs.prog_data->base.base);
    }
-
-   size_t table_size_in_bytes = entries * sizeof(uint32_t);
-
-   uint32_t *bind = brw_state_batch(brw, AUB_TRACE_BINDING_TABLE,
-                                    table_size_in_bytes, 32,
-                                    &stage_state->bind_bo_offset);
-
-   /* BRW_NEW_SURFACES and BRW_NEW_*_CONSTBUF */
-   memcpy(bind, stage_state->surf_offset, table_size_in_bytes);
-
-   brw->state.dirty.brw |= brw_new_binding_table;
 }
 
+const struct brw_tracked_state brw_vs_abo_surfaces = {
+   .dirty = {
+      .mesa = _NEW_PROGRAM,
+      .brw = BRW_NEW_ATOMIC_BUFFER |
+             BRW_NEW_BATCH |
+             BRW_NEW_BLORP |
+             BRW_NEW_VS_PROG_DATA,
+   },
+   .emit = brw_upload_vs_abo_surfaces,
+};
 
-/**
- * Constructs the binding table for the WM surface state, which maps unit
- * numbers to surface state objects.
- */
 static void
-brw_vs_upload_binding_table(struct brw_context *brw)
+brw_upload_vs_image_surfaces(struct brw_context *brw)
 {
-   struct brw_stage_state *stage_state = &brw->vs.base;
-   /* CACHE_NEW_VS_PROG */
-   const struct brw_vec4_prog_data *prog_data = &brw->vs.prog_data->base;
+   struct gl_context *ctx = &brw->ctx;
+   /* BRW_NEW_VERTEX_PROGRAM */
+   struct gl_shader_program *prog =
+      ctx->_Shader->CurrentProgram[MESA_SHADER_VERTEX];
 
-   /* BRW_NEW_SURFACES and BRW_NEW_VS_CONSTBUF */
-   brw_vec4_upload_binding_table(brw, BRW_NEW_VS_BINDING_TABLE, stage_state,
-                                 prog_data);
+   if (prog) {
+      /* BRW_NEW_VS_PROG_DATA, BRW_NEW_IMAGE_UNITS, _NEW_TEXTURE */
+      brw_upload_image_surfaces(brw, prog->_LinkedShaders[MESA_SHADER_VERTEX],
+                                &brw->vs.base, &brw->vs.prog_data->base.base);
+   }
 }
 
-const struct brw_tracked_state brw_vs_binding_table = {
+const struct brw_tracked_state brw_vs_image_surfaces = {
    .dirty = {
-      .mesa = 0,
-      .brw = (BRW_NEW_BATCH |
-             BRW_NEW_VS_CONSTBUF |
-             BRW_NEW_SURFACES),
-      .cache = CACHE_NEW_VS_PROG
+      .mesa = _NEW_TEXTURE,
+      .brw = BRW_NEW_BATCH |
+             BRW_NEW_BLORP |
+             BRW_NEW_IMAGE_UNITS |
+             BRW_NEW_VERTEX_PROGRAM |
+             BRW_NEW_VS_PROG_DATA,
    },
-   .emit = brw_vs_upload_binding_table,
+   .emit = brw_upload_vs_image_surfaces,
 };