gallium/draw: initial code to properly support llvm in the draw module
[mesa.git] / src / mesa / drivers / dri / i965 / brw_misc_state.c
index 26ec797b5fdd5abda64f47b2b514a4c250dac20b..f708ee00632972987fc53acc1e5353fd5b098fa6 100644 (file)
 
 static void upload_blend_constant_color(struct brw_context *brw)
 {
+   GLcontext *ctx = &brw->intel.ctx;
    struct brw_blend_constant_color bcc;
 
    memset(&bcc, 0, sizeof(bcc));      
    bcc.header.opcode = CMD_BLEND_CONSTANT_COLOR;
    bcc.header.length = sizeof(bcc)/4-2;
-   bcc.blend_constant_color[0] = brw->attribs.Color->BlendColor[0];
-   bcc.blend_constant_color[1] = brw->attribs.Color->BlendColor[1];
-   bcc.blend_constant_color[2] = brw->attribs.Color->BlendColor[2];
-   bcc.blend_constant_color[3] = brw->attribs.Color->BlendColor[3];
+   bcc.blend_constant_color[0] = ctx->Color.BlendColor[0];
+   bcc.blend_constant_color[1] = ctx->Color.BlendColor[1];
+   bcc.blend_constant_color[2] = ctx->Color.BlendColor[2];
+   bcc.blend_constant_color[3] = ctx->Color.BlendColor[3];
 
    BRW_CACHED_BATCH_STRUCT(brw, &bcc);
 }
@@ -65,12 +66,42 @@ static void upload_blend_constant_color(struct brw_context *brw)
 const struct brw_tracked_state brw_blend_constant_color = {
    .dirty = {
       .mesa = _NEW_COLOR,
-      .brw = 0,
+      .brw = BRW_NEW_CONTEXT,
       .cache = 0
    },
    .emit = upload_blend_constant_color
 };
 
+/* Constant single cliprect for framebuffer object or DRI2 drawing */
+static void upload_drawing_rect(struct brw_context *brw)
+{
+   struct intel_context *intel = &brw->intel;
+   GLcontext *ctx = &intel->ctx;
+
+   BEGIN_BATCH(4);
+   OUT_BATCH(_3DSTATE_DRAWRECT_INFO_I965);
+   OUT_BATCH(0); /* xmin, ymin */
+   OUT_BATCH(((ctx->DrawBuffer->Width - 1) & 0xffff) |
+           ((ctx->DrawBuffer->Height - 1) << 16));
+   OUT_BATCH(0);
+   ADVANCE_BATCH();
+}
+
+const struct brw_tracked_state brw_drawing_rect = {
+   .dirty = {
+      .mesa = _NEW_BUFFERS,
+      .brw = BRW_NEW_CONTEXT,
+      .cache = 0
+   },
+   .emit = upload_drawing_rect
+};
+
+static void prepare_binding_table_pointers(struct brw_context *brw)
+{
+   brw_add_validated_bo(brw, brw->vs.bind_bo);
+   brw_add_validated_bo(brw, brw->wm.bind_bo);
+}
+
 /**
  * Upload the binding table pointers, which point each stage's array of surface
  * state pointers.
@@ -82,13 +113,16 @@ static void upload_binding_table_pointers(struct brw_context *brw)
 {
    struct intel_context *intel = &brw->intel;
 
-   BEGIN_BATCH(6, IGNORE_CLIPRECTS);
+   BEGIN_BATCH(6);
    OUT_BATCH(CMD_BINDING_TABLE_PTRS << 16 | (6 - 2));
-   OUT_BATCH(0); /* vs */
+   if (brw->vs.bind_bo != NULL)
+      OUT_RELOC(brw->vs.bind_bo, I915_GEM_DOMAIN_SAMPLER, 0, 0); /* vs */
+   else
+      OUT_BATCH(0);
    OUT_BATCH(0); /* gs */
    OUT_BATCH(0); /* clip */
    OUT_BATCH(0); /* sf */
-   OUT_RELOC(brw->wm.bind_bo, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ, 0);
+   OUT_RELOC(brw->wm.bind_bo, I915_GEM_DOMAIN_SAMPLER, 0, 0); /* wm/ps */
    ADVANCE_BATCH();
 }
 
@@ -98,6 +132,7 @@ const struct brw_tracked_state brw_binding_table_pointers = {
       .brw = BRW_NEW_BATCH,
       .cache = CACHE_NEW_SURF_BIND,
    },
+   .prepare = prepare_binding_table_pointers,
    .emit = upload_binding_table_pointers,
 };
 
@@ -112,55 +147,44 @@ static void upload_pipelined_state_pointers(struct brw_context *brw )
 {
    struct intel_context *intel = &brw->intel;
 
-   BEGIN_BATCH(7, IGNORE_CLIPRECTS);
+   BEGIN_BATCH(7);
    OUT_BATCH(CMD_PIPELINED_STATE_POINTERS << 16 | (7 - 2));
-   OUT_RELOC(brw->vs.state_bo, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ, 0);
+   OUT_RELOC(brw->vs.state_bo, I915_GEM_DOMAIN_INSTRUCTION, 0, 0);
    if (brw->gs.prog_active)
-      OUT_RELOC(brw->gs.state_bo, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ, 1);
-   else
-      OUT_BATCH(0);
-   if (!brw->metaops.active)
-      OUT_RELOC(brw->clip.state_bo, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ, 1);
+      OUT_RELOC(brw->gs.state_bo, I915_GEM_DOMAIN_INSTRUCTION, 0, 1);
    else
       OUT_BATCH(0);
-   OUT_RELOC(brw->sf.state_bo, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ, 0);
-   OUT_RELOC(brw->wm.state_bo, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ, 0);
-   OUT_RELOC(brw->cc.state_bo, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ, 0);
+   OUT_RELOC(brw->clip.state_bo, I915_GEM_DOMAIN_INSTRUCTION, 0, 1);
+   OUT_RELOC(brw->sf.state_bo, I915_GEM_DOMAIN_INSTRUCTION, 0, 0);
+   OUT_RELOC(brw->wm.state_bo, I915_GEM_DOMAIN_INSTRUCTION, 0, 0);
+   OUT_RELOC(brw->cc.state_bo, I915_GEM_DOMAIN_INSTRUCTION, 0, 0);
    ADVANCE_BATCH();
 
    brw->state.dirty.brw |= BRW_NEW_PSP;
 }
 
-#if 0
-/* Combined into brw_psp_urb_cbs */
-const struct brw_tracked_state brw_pipelined_state_pointers = {
-   .dirty = {
-      .mesa = 0,
-      .brw = BRW_NEW_METAOPS | BRW_NEW_BATCH,
-      .cache = (CACHE_NEW_VS_UNIT | 
-               CACHE_NEW_GS_UNIT | 
-               CACHE_NEW_GS_PROG | 
-               CACHE_NEW_CLIP_UNIT | 
-               CACHE_NEW_SF_UNIT | 
-               CACHE_NEW_WM_UNIT | 
-               CACHE_NEW_CC_UNIT)
-   },
-   .emit = upload_pipelined_state_pointers
-};
-#endif
+
+static void prepare_psp_urb_cbs(struct brw_context *brw)
+{
+   brw_add_validated_bo(brw, brw->vs.state_bo);
+   brw_add_validated_bo(brw, brw->gs.state_bo);
+   brw_add_validated_bo(brw, brw->clip.state_bo);
+   brw_add_validated_bo(brw, brw->sf.state_bo);
+   brw_add_validated_bo(brw, brw->wm.state_bo);
+   brw_add_validated_bo(brw, brw->cc.state_bo);
+}
 
 static void upload_psp_urb_cbs(struct brw_context *brw )
 {
    upload_pipelined_state_pointers(brw);
    brw_upload_urb_fence(brw);
-   brw_upload_constant_buffer_state(brw);
+   brw_upload_cs_urb_state(brw);
 }
 
-
 const struct brw_tracked_state brw_psp_urb_cbs = {
    .dirty = {
       .mesa = 0,
-      .brw = BRW_NEW_URB_FENCE | BRW_NEW_METAOPS | BRW_NEW_BATCH,
+      .brw = BRW_NEW_URB_FENCE | BRW_NEW_BATCH,
       .cache = (CACHE_NEW_VS_UNIT | 
                CACHE_NEW_GS_UNIT | 
                CACHE_NEW_GS_PROG | 
@@ -169,33 +193,26 @@ const struct brw_tracked_state brw_psp_urb_cbs = {
                CACHE_NEW_WM_UNIT | 
                CACHE_NEW_CC_UNIT)
    },
+   .prepare = prepare_psp_urb_cbs,
    .emit = upload_psp_urb_cbs,
 };
 
-/**
- * Upload the depthbuffer offset and format.
- *
- * We have to do this per state validation as we need to emit the relocation
- * in the batch buffer.
- */
-
-static int prepare_depthbuffer(struct brw_context *brw)
+static void prepare_depthbuffer(struct brw_context *brw)
 {
    struct intel_region *region = brw->state.depth_region;
 
-   if (!region || !region->buffer)
-      return 0;
-   return dri_bufmgr_check_aperture_space(region->buffer);
+   if (region != NULL)
+      brw_add_validated_bo(brw, region->buffer);
 }
 
 static void emit_depthbuffer(struct brw_context *brw)
 {
    struct intel_context *intel = &brw->intel;
    struct intel_region *region = brw->state.depth_region;
-   unsigned int len = BRW_IS_IGD(brw) ? sizeof(struct brw_depthbuffer_igd) / 4 : sizeof(struct brw_depthbuffer) / 4;
+   unsigned int len = (intel->is_g4x || intel->is_ironlake) ? 6 : 5;
 
    if (region == NULL) {
-      BEGIN_BATCH(len, IGNORE_CLIPRECTS);
+      BEGIN_BATCH(len);
       OUT_BATCH(CMD_DEPTH_BUFFER << 16 | (len - 2));
       OUT_BATCH((BRW_DEPTHFORMAT_D32_FLOAT << 18) |
                (BRW_SURFACE_NULL << 29));
@@ -203,7 +220,7 @@ static void emit_depthbuffer(struct brw_context *brw)
       OUT_BATCH(0);
       OUT_BATCH(0);
 
-      if (BRW_IS_IGD(brw))
+      if (intel->is_g4x || intel->is_ironlake)
          OUT_BATCH(0);
 
       ADVANCE_BATCH();
@@ -225,21 +242,24 @@ static void emit_depthbuffer(struct brw_context *brw)
         return;
       }
 
-      BEGIN_BATCH(len, IGNORE_CLIPRECTS);
+      assert(region->tiling != I915_TILING_X);
+
+      BEGIN_BATCH(len);
       OUT_BATCH(CMD_DEPTH_BUFFER << 16 | (len - 2));
       OUT_BATCH(((region->pitch * region->cpp) - 1) |
                (format << 18) |
                (BRW_TILEWALK_YMAJOR << 26) |
-               (region->tiled << 27) |
+               ((region->tiling != I915_TILING_NONE) << 27) |
                (BRW_SURFACE_2D << 29));
       OUT_RELOC(region->buffer,
-               DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ | DRM_BO_FLAG_WRITE, 0);
+               I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
+               0);
       OUT_BATCH((BRW_SURFACE_MIPMAPLAYOUT_BELOW << 1) |
                ((region->pitch - 1) << 6) |
                ((region->height - 1) << 19));
       OUT_BATCH(0);
 
-      if (BRW_IS_IGD(brw))
+      if (intel->is_g4x || intel->is_ironlake)
          OUT_BATCH(0);
 
       ADVANCE_BATCH();
@@ -264,6 +284,7 @@ const struct brw_tracked_state brw_depthbuffer = {
 
 static void upload_polygon_stipple(struct brw_context *brw)
 {
+   GLcontext *ctx = &brw->intel.ctx;
    struct brw_polygon_stipple bps;
    GLuint i;
 
@@ -271,8 +292,21 @@ static void upload_polygon_stipple(struct brw_context *brw)
    bps.header.opcode = CMD_POLY_STIPPLE_PATTERN;
    bps.header.length = sizeof(bps)/4-2;
 
-   for (i = 0; i < 32; i++)
-      bps.stipple[i] = brw->attribs.PolygonStipple[31 - i]; /* invert */
+   /* Polygon stipple is provided in OpenGL order, i.e. bottom
+    * row first.  If we're rendering to a window (i.e. the
+    * default frame buffer object, 0), then we need to invert
+    * it to match our pixel layout.  But if we're rendering
+    * to a FBO (i.e. any named frame buffer object), we *don't*
+    * need to invert - we already match the layout.
+    */
+   if (ctx->DrawBuffer->Name == 0) {
+      for (i = 0; i < 32; i++)
+         bps.stipple[i] = ctx->PolygonStipple[31 - i]; /* invert */
+   }
+   else {
+      for (i = 0; i < 32; i++)
+         bps.stipple[i] = ctx->PolygonStipple[i]; /* don't invert */
+   }
 
    BRW_CACHED_BATCH_STRUCT(brw, &bps);
 }
@@ -280,7 +314,7 @@ static void upload_polygon_stipple(struct brw_context *brw)
 const struct brw_tracked_state brw_polygon_stipple = {
    .dirty = {
       .mesa = _NEW_POLYGONSTIPPLE,
-      .brw = 0,
+      .brw = BRW_NEW_CONTEXT,
       .cache = 0
    },
    .emit = upload_polygon_stipple
@@ -293,15 +327,29 @@ const struct brw_tracked_state brw_polygon_stipple = {
 
 static void upload_polygon_stipple_offset(struct brw_context *brw)
 {
-   __DRIdrawablePrivate *dPriv = brw->intel.driDrawable;
+   GLcontext *ctx = &brw->intel.ctx;
    struct brw_polygon_stipple_offset bpso;
 
    memset(&bpso, 0, sizeof(bpso));
    bpso.header.opcode = CMD_POLY_STIPPLE_OFFSET;
    bpso.header.length = sizeof(bpso)/4-2;
 
-   bpso.bits0.x_offset = (32 - (dPriv->x & 31)) & 31;
-   bpso.bits0.y_offset = (32 - ((dPriv->y + dPriv->h) & 31)) & 31;
+   /* If we're drawing to a system window (ctx->DrawBuffer->Name == 0),
+    * we have to invert the Y axis in order to match the OpenGL
+    * pixel coordinate system, and our offset must be matched
+    * to the window position.  If we're drawing to a FBO
+    * (ctx->DrawBuffer->Name != 0), then our native pixel coordinate
+    * system works just fine, and there's no window system to
+    * worry about.
+    */
+   if (brw->intel.ctx.DrawBuffer->Name == 0) {
+      bpso.bits0.x_offset = 0;
+      bpso.bits0.y_offset = (32 - (ctx->DrawBuffer->Height & 31)) & 31;
+   }
+   else {
+      bpso.bits0.y_offset = 0;
+      bpso.bits0.x_offset = 0;
+   }
 
    BRW_CACHED_BATCH_STRUCT(brw, &bpso);
 }
@@ -311,7 +359,7 @@ static void upload_polygon_stipple_offset(struct brw_context *brw)
 const struct brw_tracked_state brw_polygon_stipple_offset = {
    .dirty = {
       .mesa = _NEW_WINDOW_POS,
-      .brw = 0,
+      .brw = BRW_NEW_CONTEXT,
       .cache = 0
    },
    .emit = upload_polygon_stipple_offset
@@ -323,8 +371,8 @@ const struct brw_tracked_state brw_polygon_stipple_offset = {
 static void upload_aa_line_parameters(struct brw_context *brw)
 {
    struct brw_aa_line_parameters balp;
-   
-   if (!BRW_IS_IGD(brw))
+
+   if (!brw->has_aa_line_parameters)
       return;
 
    /* use legacy aa line coverage computation */
@@ -350,6 +398,7 @@ const struct brw_tracked_state brw_aa_line_parameters = {
 
 static void upload_line_stipple(struct brw_context *brw)
 {
+   GLcontext *ctx = &brw->intel.ctx;
    struct brw_line_stipple bls;
    GLfloat tmp;
    GLint tmpi;
@@ -358,10 +407,10 @@ static void upload_line_stipple(struct brw_context *brw)
    bls.header.opcode = CMD_LINE_STIPPLE_PATTERN;
    bls.header.length = sizeof(bls)/4 - 2;
 
-   bls.bits0.pattern = brw->attribs.Line->StipplePattern;
-   bls.bits1.repeat_count = brw->attribs.Line->StippleFactor;
+   bls.bits0.pattern = ctx->Line.StipplePattern;
+   bls.bits1.repeat_count = ctx->Line.StippleFactor;
 
-   tmp = 1.0 / (GLfloat) brw->attribs.Line->StippleFactor;
+   tmp = 1.0 / (GLfloat) ctx->Line.StippleFactor;
    tmpi = tmp * (1<<13);
 
 
@@ -373,47 +422,13 @@ static void upload_line_stipple(struct brw_context *brw)
 const struct brw_tracked_state brw_line_stipple = {
    .dirty = {
       .mesa = _NEW_LINE,
-      .brw = 0,
+      .brw = BRW_NEW_CONTEXT,
       .cache = 0
    },
    .emit = upload_line_stipple
 };
 
 
-
-/***********************************************************************
- * Misc constant state packets
- */
-
-static void upload_pipe_control(struct brw_context *brw)
-{
-   struct brw_pipe_control pc;
-
-   return;
-
-   memset(&pc, 0, sizeof(pc));
-
-   pc.header.opcode = CMD_PIPE_CONTROL;
-   pc.header.length = sizeof(pc)/4 - 2;
-   pc.header.post_sync_operation = PIPE_CONTROL_NOWRITE;
-
-   pc.header.instruction_state_cache_flush_enable = 1;
-
-   pc.bits1.dest_addr_type = PIPE_CONTROL_GTTWRITE_GLOBAL;
-
-   BRW_BATCH_STRUCT(brw, &pc);
-}
-
-const struct brw_tracked_state brw_pipe_control = {
-   .dirty = {
-      .mesa = 0,
-      .brw = BRW_NEW_BATCH,
-      .cache = 0
-   },
-   .emit = upload_pipe_control
-};
-
-
 /***********************************************************************
  * Misc invarient state packets
  */
@@ -426,7 +441,7 @@ static void upload_invarient_state( struct brw_context *brw )
       struct brw_pipeline_select ps;
 
       memset(&ps, 0, sizeof(ps));
-      ps.header.opcode = CMD_PIPELINE_SELECT(brw);
+      ps.header.opcode = brw->CMD_PIPELINE_SELECT;
       ps.header.pipeline_select = 0;
       BRW_BATCH_STRUCT(brw, &ps);
    }
@@ -462,7 +477,7 @@ static void upload_invarient_state( struct brw_context *brw )
       struct brw_vf_statistics vfs;
       memset(&vfs, 0, sizeof(vfs));
 
-      vfs.opcode = CMD_VF_STATISTICS(brw);
+      vfs.opcode = brw->CMD_VF_STATISTICS;
       if (INTEL_DEBUG & DEBUG_STATS)
         vfs.statistics_enable = 1; 
 
@@ -494,14 +509,27 @@ static void upload_state_base_address( struct brw_context *brw )
    /* Output the structure (brw_state_base_address) directly to the
     * batchbuffer, so we can emit relocations inline.
     */
-   BEGIN_BATCH(6, IGNORE_CLIPRECTS);
-   OUT_BATCH(CMD_STATE_BASE_ADDRESS << 16 | (6 - 2));
-   OUT_BATCH(1); /* General state base address */
-   OUT_BATCH(1); /* Surface state base address */
-   OUT_BATCH(1); /* Indirect object base address */
-   OUT_BATCH(1); /* General state upper bound */
-   OUT_BATCH(1); /* Indirect object upper bound */
-   ADVANCE_BATCH();
+   if (intel->is_ironlake) {
+       BEGIN_BATCH(8);
+       OUT_BATCH(CMD_STATE_BASE_ADDRESS << 16 | (8 - 2));
+       OUT_BATCH(1); /* General state base address */
+       OUT_BATCH(1); /* Surface state base address */
+       OUT_BATCH(1); /* Indirect object base address */
+       OUT_BATCH(1); /* Instruction base address */
+       OUT_BATCH(1); /* General state upper bound */
+       OUT_BATCH(1); /* Indirect object upper bound */
+       OUT_BATCH(1); /* Instruction access upper bound */
+       ADVANCE_BATCH();
+   } else {
+       BEGIN_BATCH(6);
+       OUT_BATCH(CMD_STATE_BASE_ADDRESS << 16 | (6 - 2));
+       OUT_BATCH(1); /* General state base address */
+       OUT_BATCH(1); /* Surface state base address */
+       OUT_BATCH(1); /* Indirect object base address */
+       OUT_BATCH(1); /* General state upper bound */
+       OUT_BATCH(1); /* Indirect object upper bound */
+       ADVANCE_BATCH();
+   }
 }
 
 const struct brw_tracked_state brw_state_base_address = {