i965/fs: Set default access mode to Align1 for all instructions in the generator.
[mesa.git] / src / mesa / drivers / dri / i965 / gen7_sf_state.c
index 7691cb218d69807bcd0918ae495eca50575b0d14..7a3cc530c0aea9f5c430622893a113a53a7d3515 100644 (file)
 #include "brw_defines.h"
 #include "brw_util.h"
 #include "main/macros.h"
+#include "main/fbobject.h"
+#include "main/framebuffer.h"
 #include "intel_batchbuffer.h"
 
 static void
 upload_sbe_state(struct brw_context *brw)
 {
-   struct intel_context *intel = &brw->intel;
-   struct gl_context *ctx = &intel->ctx;
-   struct brw_vue_map vue_map;
-   uint32_t urb_entry_read_length;
-   /* CACHE_NEW_VS_PROG */
-   GLbitfield64 vs_outputs_written = brw->vs.prog_data->outputs_written;
-   /* BRW_NEW_FRAGMENT_PROGRAM */
-   uint32_t num_outputs = _mesa_bitcount_64(brw->fragment_program->Base.InputsRead);
-   /* _NEW_LIGHT */
-   bool shade_model_flat = ctx->Light.ShadeModel == GL_FLAT;
-   uint32_t dw1, dw10, dw11;
+   struct gl_context *ctx = &brw->ctx;
+   /* BRW_NEW_FS_PROG_DATA */
+   uint32_t num_outputs = brw->wm.prog_data->num_varying_inputs;
+   uint32_t dw1;
+   uint32_t point_sprite_enables;
    int i;
-   int attr = 0, input_index = 0;
-   /* _NEW_TRANSFORM */
-   int urb_entry_read_offset = 1;
-   bool userclip_active = (ctx->Transform.ClipPlanesEnabled != 0);
-   uint16_t attr_overrides[FRAG_ATTRIB_MAX];
-
-   brw_compute_vue_map(&vue_map, intel, userclip_active, vs_outputs_written);
-   urb_entry_read_length = (vue_map.num_slots + 1)/2 - urb_entry_read_offset;
-   if (urb_entry_read_length == 0) {
-      /* Setting the URB entry read length to 0 causes undefined behavior, so
-       * if we have no URB data to read, set it to 1.
-       */
-      urb_entry_read_length = 1;
-   }
+   uint16_t attr_overrides[16];
+   /* _NEW_BUFFERS */
+   bool render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer);
+   uint32_t point_sprite_origin;
 
    /* FINISHME: Attribute Swizzle Control Mode? */
-   dw1 =
-      GEN7_SBE_SWIZZLE_ENABLE |
-      num_outputs << GEN7_SBE_NUM_OUTPUTS_SHIFT |
-      urb_entry_read_length << GEN7_SBE_URB_ENTRY_READ_LENGTH_SHIFT |
-      urb_entry_read_offset << GEN7_SBE_URB_ENTRY_READ_OFFSET_SHIFT;
-
-   /* _NEW_POINT */
-   if (ctx->Point.SpriteOrigin == GL_LOWER_LEFT)
-      dw1 |= GEN6_SF_POINT_SPRITE_LOWERLEFT;
+   dw1 = GEN7_SBE_SWIZZLE_ENABLE | num_outputs << GEN7_SBE_NUM_OUTPUTS_SHIFT;
 
-   dw10 = 0;
-   dw11 = 0;
-
-   /* Create the mapping from the FS inputs we produce to the VS outputs
-    * they source from.
+   /* _NEW_POINT
+    *
+    * Window coordinates in an FBO are inverted, which means point
+    * sprite origin must be inverted.
     */
-   for (; attr < FRAG_ATTRIB_MAX; attr++) {
-      enum glsl_interp_qualifier interp_qualifier =
-         brw->fragment_program->InterpQualifier[attr];
-      bool is_gl_Color = attr == FRAG_ATTRIB_COL0 || attr == FRAG_ATTRIB_COL1;
-
-      if (!(brw->fragment_program->Base.InputsRead & BITFIELD64_BIT(attr)))
-        continue;
-
-      if (ctx->Point.PointSprite &&
-         attr >= FRAG_ATTRIB_TEX0 && attr <= FRAG_ATTRIB_TEX7 &&
-         ctx->Point.CoordReplace[attr - FRAG_ATTRIB_TEX0]) {
-        dw10 |= (1 << input_index);
-      }
-
-      if (attr == FRAG_ATTRIB_PNTC)
-        dw10 |= (1 << input_index);
-
-      /* flat shading */
-      if (interp_qualifier == INTERP_QUALIFIER_FLAT ||
-          (shade_model_flat && is_gl_Color &&
-           interp_qualifier == INTERP_QUALIFIER_NONE))
-         dw11 |= (1 << input_index);
-
-      /* The hardware can only do the overrides on 16 overrides at a
-       * time, and the other up to 16 have to be lined up so that the
-       * input index = the output index.  We'll need to do some
-       * tweaking to make sure that's the case.
-       */
-      assert(input_index < 16 || attr == input_index);
-
-      /* _NEW_LIGHT | _NEW_PROGRAM */
-      attr_overrides[input_index++] =
-         get_attr_override(&vue_map, urb_entry_read_offset, attr,
-                           ctx->VertexProgram._TwoSideEnabled);
+   if ((ctx->Point.SpriteOrigin == GL_LOWER_LEFT) != render_to_fbo) {
+      point_sprite_origin = GEN6_SF_POINT_SPRITE_LOWERLEFT;
+   } else {
+      point_sprite_origin = GEN6_SF_POINT_SPRITE_UPPERLEFT;
    }
+   dw1 |= point_sprite_origin;
 
-   for (; input_index < FRAG_ATTRIB_MAX; input_index++)
-      attr_overrides[input_index] = 0;
+   /* BRW_NEW_VUE_MAP_GEOM_OUT | BRW_NEW_FRAGMENT_PROGRAM
+    * _NEW_POINT | _NEW_LIGHT | _NEW_PROGRAM | BRW_NEW_FS_PROG_DATA
+    */
+   uint32_t urb_entry_read_length;
+   uint32_t urb_entry_read_offset;
+   calculate_attr_overrides(brw, attr_overrides, &point_sprite_enables,
+                            &urb_entry_read_length, &urb_entry_read_offset);
+   dw1 |= urb_entry_read_length << GEN7_SBE_URB_ENTRY_READ_LENGTH_SHIFT |
+          urb_entry_read_offset << GEN7_SBE_URB_ENTRY_READ_OFFSET_SHIFT;
 
    BEGIN_BATCH(14);
    OUT_BATCH(_3DSTATE_SBE << 16 | (14 - 2));
@@ -123,8 +78,8 @@ upload_sbe_state(struct brw_context *brw)
       OUT_BATCH(attr_overrides[i * 2] | attr_overrides[i * 2 + 1] << 16);
    }
 
-   OUT_BATCH(dw10); /* point sprite texcoord bitmask */
-   OUT_BATCH(dw11); /* constant interp bitmask */
+   OUT_BATCH(point_sprite_enables); /* dw10 */
+   OUT_BATCH(brw->wm.prog_data->flat_inputs);
    OUT_BATCH(0); /* wrapshortest enables 0-7 */
    OUT_BATCH(0); /* wrapshortest enables 8-15 */
    ADVANCE_BATCH();
@@ -132,14 +87,17 @@ upload_sbe_state(struct brw_context *brw)
 
 const struct brw_tracked_state gen7_sbe_state = {
    .dirty = {
-      .mesa  = (_NEW_LIGHT |
-               _NEW_POINT |
-               _NEW_PROGRAM |
-               _NEW_TRANSFORM),
-      .brw   = (BRW_NEW_CONTEXT |
-               BRW_NEW_FRAGMENT_PROGRAM |
-               BRW_NEW_HIZ),
-      .cache = CACHE_NEW_VS_PROG
+      .mesa  = _NEW_BUFFERS |
+               _NEW_LIGHT |
+               _NEW_POINT |
+               _NEW_PROGRAM,
+      .brw   = BRW_NEW_BLORP |
+               BRW_NEW_CONTEXT |
+               BRW_NEW_FRAGMENT_PROGRAM |
+               BRW_NEW_FS_PROG_DATA |
+               BRW_NEW_GEOMETRY_PROGRAM |
+               BRW_NEW_PRIMITIVE |
+               BRW_NEW_VUE_MAP_GEOM_OUT,
    },
    .emit = upload_sbe_state,
 };
@@ -147,30 +105,23 @@ const struct brw_tracked_state gen7_sbe_state = {
 static void
 upload_sf_state(struct brw_context *brw)
 {
-   struct intel_context *intel = &brw->intel;
-   struct gl_context *ctx = &intel->ctx;
+   struct gl_context *ctx = &brw->ctx;
    uint32_t dw1, dw2, dw3;
    float point_size;
    /* _NEW_BUFFERS */
-   bool render_to_fbo = brw->intel.ctx.DrawBuffer->Name != 0;
+   bool render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer);
+   const bool multisampled_fbo = _mesa_geometric_samples(ctx->DrawBuffer) > 1;
 
    dw1 = GEN6_SF_STATISTICS_ENABLE;
 
-   /* Enable viewport transform only if no HiZ operation is progress
-    *
-    * From page 11 of the SandyBridge PRM, Volume 2, Part 1, Section 1.3, "3D
-    * Primitives Overview":
-    *     RECTLIST: Viewport Mapping must be DISABLED (as is typical with the
-    *     use of screen- space coordinates).
-    */
-   if (!brw->hiz.op)
-      dw1 |= GEN6_SF_VIEWPORT_TRANSFORM_ENABLE;
+   if (brw->sf.viewport_transform_enable)
+       dw1 |= GEN6_SF_VIEWPORT_TRANSFORM_ENABLE;
 
    /* _NEW_BUFFERS */
    dw1 |= (brw_depthbuffer_format(brw) << GEN7_SF_DEPTH_BUFFER_SURFACE_FORMAT_SHIFT);
 
    /* _NEW_POLYGON */
-   if ((ctx->Polygon.FrontFace == GL_CCW) ^ render_to_fbo)
+   if (ctx->Polygon._FrontBit == render_to_fbo)
       dw1 |= GEN6_SF_WINDING_CCW;
 
    if (ctx->Polygon.OffsetFill)
@@ -196,8 +147,7 @@ upload_sf_state(struct brw_context *brw)
        break;
 
    default:
-       assert(0);
-       break;
+       unreachable("not reached");
    }
 
    switch (ctx->Polygon.BackMode) {
@@ -214,8 +164,7 @@ upload_sf_state(struct brw_context *brw)
        break;
 
    default:
-       assert(0);
-       break;
+       unreachable("not reached");
    }
 
    dw2 = 0;
@@ -232,31 +181,37 @@ upload_sf_state(struct brw_context *brw)
         dw2 |= GEN6_SF_CULL_BOTH;
         break;
       default:
-        assert(0);
-        break;
+        unreachable("not reached");
       }
    } else {
       dw2 |= GEN6_SF_CULL_NONE;
    }
 
-   /* _NEW_SCISSOR */
-   if (ctx->Scissor.Enabled)
+   /* _NEW_SCISSOR _NEW_POLYGON BRW_NEW_GEOMETRY_PROGRAM BRW_NEW_PRIMITIVE */
+   if (ctx->Scissor.EnableFlags ||
+       is_drawing_points(brw) || is_drawing_lines(brw))
       dw2 |= GEN6_SF_SCISSOR_ENABLE;
 
    /* _NEW_LINE */
-   dw2 |= U_FIXED(CLAMP(ctx->Line.Width, 0.0, 7.99), 7) <<
-      GEN6_SF_LINE_WIDTH_SHIFT;
+   {
+      uint32_t line_width_u3_7 = brw_get_line_width(brw);
+      dw2 |= line_width_u3_7 << GEN6_SF_LINE_WIDTH_SHIFT;
+   }
    if (ctx->Line.SmoothFlag) {
       dw2 |= GEN6_SF_LINE_AA_ENABLE;
-      dw2 |= GEN6_SF_LINE_AA_MODE_TRUE;
       dw2 |= GEN6_SF_LINE_END_CAP_WIDTH_1_0;
    }
+   if (ctx->Line.StippleFlag && brw->is_haswell) {
+      dw2 |= HSW_SF_LINE_STIPPLE_ENABLE;
+   }
+   /* _NEW_MULTISAMPLE */
+   if (multisampled_fbo && ctx->Multisample.Enabled)
+      dw2 |= GEN6_SF_MSRAST_ON_PATTERN;
 
    /* FINISHME: Last Pixel Enable?  Vertex Sub Pixel Precision Select?
-    * FINISHME: AA Line Distance Mode?
     */
 
-   dw3 = 0;
+   dw3 = GEN6_SF_LINE_AA_MODE_TRUE;
 
    /* _NEW_PROGRAM | _NEW_POINT */
    if (!(ctx->VertexProgram.PointSizeEnabled || ctx->Point._Attenuated))
@@ -266,7 +221,7 @@ upload_sf_state(struct brw_context *brw)
    point_size = CLAMP(ctx->Point.Size, ctx->Point.MinSize, ctx->Point.MaxSize);
 
    /* Clamp to the hardware limits and convert to fixed point */
-   dw3 |= U_FIXED(CLAMP(point_size, 0.125, 255.875), 3);
+   dw3 |= U_FIXED(CLAMP(point_size, 0.125f, 255.875f), 3);
 
    /* _NEW_LIGHT */
    if (ctx->Light.ProvokingVertex != GL_FIRST_VERTEX_CONVENTION) {
@@ -285,22 +240,23 @@ upload_sf_state(struct brw_context *brw)
    OUT_BATCH(dw3);
    OUT_BATCH_F(ctx->Polygon.OffsetUnits * 2); /* constant.  copied from gen4 */
    OUT_BATCH_F(ctx->Polygon.OffsetFactor); /* scale */
-   OUT_BATCH_F(0.0); /* XXX: global depth offset clamp */
+   OUT_BATCH_F(ctx->Polygon.OffsetClamp); /* global depth offset clamp */
    ADVANCE_BATCH();
 }
 
 const struct brw_tracked_state gen7_sf_state = {
    .dirty = {
-      .mesa  = (_NEW_LIGHT |
-               _NEW_PROGRAM |
-               _NEW_POLYGON |
-               _NEW_LINE |
-               _NEW_SCISSOR |
-               _NEW_BUFFERS |
-               _NEW_POINT),
-      .brw   = (BRW_NEW_CONTEXT |
-               BRW_NEW_HIZ),
-      .cache = CACHE_NEW_VS_PROG
+      .mesa  = _NEW_BUFFERS |
+               _NEW_LIGHT |
+               _NEW_LINE |
+               _NEW_MULTISAMPLE |
+               _NEW_POINT |
+               _NEW_POLYGON |
+               _NEW_PROGRAM |
+               _NEW_SCISSOR,
+      .brw   = BRW_NEW_BLORP |
+               BRW_NEW_CONTEXT |
+               BRW_NEW_PRIMITIVE,
    },
    .emit = upload_sf_state,
 };