#include "main/macros.h"
#include "intel_batchbuffer.h"
-static uint32_t
-get_attr_override(struct brw_context *brw, int fs_attr, int two_side_color)
+/**
+ * Determine the appropriate attribute override value to store into the
+ * 3DSTATE_SF structure for a given fragment shader attribute. The attribute
+ * override value contains two pieces of information: the location of the
+ * attribute in the VUE (relative to urb_entry_read_offset, see below), and a
+ * flag indicating whether to "swizzle" the attribute based on the direction
+ * the triangle is facing.
+ *
+ * If an attribute is "swizzled", then the given VUE location is used for
+ * front-facing triangles, and the VUE location that immediately follows is
+ * used for back-facing triangles. We use this to implement the mapping from
+ * gl_FrontColor/gl_BackColor to gl_Color.
+ *
+ * urb_entry_read_offset is the offset into the VUE at which the SF unit is
+ * being instructed to begin reading attribute data. It can be set to a
+ * nonzero value to prevent the SF unit from wasting time reading elements of
+ * the VUE that are not needed by the fragment shader. It is measured in
+ * 256-bit increments.
+ */
+uint32_t
+get_attr_override(struct brw_vue_map *vue_map, int urb_entry_read_offset,
+ int fs_attr, bool two_side_color)
{
- int attr_index = 0, i, vs_attr;
- int bfc = 0;
-
- if (fs_attr <= FRAG_ATTRIB_TEX7)
- vs_attr = fs_attr;
- else if (fs_attr == FRAG_ATTRIB_FACE)
- vs_attr = 0; /* XXX */
- else if (fs_attr == FRAG_ATTRIB_PNTC)
- vs_attr = 0; /* XXX */
- else {
- assert(fs_attr >= FRAG_ATTRIB_VAR0);
- vs_attr = fs_attr - FRAG_ATTRIB_VAR0 + VERT_RESULT_VAR0;
+ int attr_override, slot;
+ int vs_attr = _mesa_frag_attrib_to_vert_result(fs_attr);
+ if (vs_attr < 0 || vs_attr == VERT_RESULT_HPOS) {
+ /* These attributes will be overwritten by the fragment shader's
+ * interpolation code (see emit_interp() in brw_wm_fp.c), so just let
+ * them reference the first available attribute.
+ */
+ return 0;
}
- /* Find the source index (0 = first attribute after the 4D position)
- * for this output attribute. attr is currently a VERT_RESULT_* but should
- * be FRAG_ATTRIB_*.
+ /* Find the VUE slot for this attribute. */
+ slot = vue_map->vert_result_to_slot[vs_attr];
+
+ /* If there was only a back color written but not front, use back
+ * as the color instead of undefined
*/
- for (i = 1; i < vs_attr; i++) {
- if (brw->vs.prog_data->outputs_written & BITFIELD64_BIT(i))
- attr_index++;
+ if (slot == -1 && vs_attr == VERT_RESULT_COL0)
+ slot = vue_map->vert_result_to_slot[VERT_RESULT_BFC0];
+ if (slot == -1 && vs_attr == VERT_RESULT_COL1)
+ slot = vue_map->vert_result_to_slot[VERT_RESULT_BFC1];
+
+ if (slot == -1) {
+ /* This attribute does not exist in the VUE--that means that the vertex
+ * shader did not write to it. Behavior is undefined in this case, so
+ * just reference the first available attribute.
+ */
+ return 0;
}
- assert(attr_index < 32);
+ /* Compute the location of the attribute relative to urb_entry_read_offset.
+ * Each increment of urb_entry_read_offset represents a 256-bit value, so
+ * it counts for two 128-bit VUE slots.
+ */
+ attr_override = slot - 2 * urb_entry_read_offset;
+ assert (attr_override >= 0 && attr_override < 32);
+ /* If we are doing two-sided color, and the VUE slot following this one
+ * represents a back-facing color, then we need to instruct the SF unit to
+ * do back-facing swizzling.
+ */
if (two_side_color) {
- if ((brw->vs.prog_data->outputs_written & BITFIELD64_BIT(VERT_RESULT_COL1)) &&
- (brw->vs.prog_data->outputs_written & BITFIELD64_BIT(VERT_RESULT_BFC1))) {
- assert(brw->vs.prog_data->outputs_written & BITFIELD64_BIT(VERT_RESULT_COL0));
- assert(brw->vs.prog_data->outputs_written & BITFIELD64_BIT(VERT_RESULT_BFC0));
- bfc = 2;
- } else if ((brw->vs.prog_data->outputs_written & BITFIELD64_BIT(VERT_RESULT_COL0)) &&
- (brw->vs.prog_data->outputs_written & BITFIELD64_BIT(VERT_RESULT_BFC0)))
- bfc = 1;
+ if (vue_map->slot_to_vert_result[slot] == VERT_RESULT_COL0 &&
+ vue_map->slot_to_vert_result[slot+1] == VERT_RESULT_BFC0)
+ attr_override |= (ATTRIBUTE_SWIZZLE_INPUTATTR_FACING << ATTRIBUTE_SWIZZLE_SHIFT);
+ else if (vue_map->slot_to_vert_result[slot] == VERT_RESULT_COL1 &&
+ vue_map->slot_to_vert_result[slot+1] == VERT_RESULT_BFC1)
+ attr_override |= (ATTRIBUTE_SWIZZLE_INPUTATTR_FACING << ATTRIBUTE_SWIZZLE_SHIFT);
}
- if (bfc && (fs_attr <= FRAG_ATTRIB_TEX7 && fs_attr > FRAG_ATTRIB_WPOS)) {
- if (fs_attr == FRAG_ATTRIB_COL0)
- attr_index |= (ATTRIBUTE_SWIZZLE_INPUTATTR_FACING << ATTRIBUTE_SWIZZLE_SHIFT);
- else if (fs_attr == FRAG_ATTRIB_COL1 && bfc == 2) {
- attr_index++;
- attr_index |= (ATTRIBUTE_SWIZZLE_INPUTATTR_FACING << ATTRIBUTE_SWIZZLE_SHIFT);
- } else {
- attr_index += bfc;
- }
- }
-
- return attr_index;
+ return attr_override;
}
static void
{
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 */
- uint32_t num_inputs = brw_count_bits(brw->vs.prog_data->outputs_written);
+ GLbitfield64 vs_outputs_written = brw->vs.prog_data->outputs_written;
/* BRW_NEW_FRAGMENT_PROGRAM */
- uint32_t num_outputs = brw_count_bits(brw->fragment_program->Base.InputsRead);
+ 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, dw2, dw3, dw4, dw16, dw17;
int i;
/* _NEW_BUFFER */
- GLboolean render_to_fbo = brw->intel.ctx.DrawBuffer->Name != 0;
- int attr = 0;
- int urb_start;
- int two_side_color = (ctx->Light.Enabled && ctx->Light.Model.TwoSide);
+ bool render_to_fbo = brw->intel.ctx.DrawBuffer->Name != 0;
+ int attr = 0, input_index = 0;
+ int urb_entry_read_offset = 1;
+ float point_size;
+ uint16_t attr_overrides[FRAG_ATTRIB_MAX];
+ bool userclip_active;
/* _NEW_TRANSFORM */
- if (ctx->Transform.ClipPlanesEnabled)
- urb_start = 2;
- else
- urb_start = 1;
+ userclip_active = (ctx->Transform.ClipPlanesEnabled != 0);
+
+ 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;
+ }
dw1 =
GEN6_SF_SWIZZLE_ENABLE |
num_outputs << GEN6_SF_NUM_OUTPUTS_SHIFT |
- (num_inputs + 1) / 2 << GEN6_SF_URB_ENTRY_READ_LENGTH_SHIFT |
- urb_start << GEN6_SF_URB_ENTRY_READ_OFFSET_SHIFT;
- dw2 = GEN6_SF_VIEWPORT_TRANSFORM_ENABLE |
- GEN6_SF_STATISTICS_ENABLE;
+ urb_entry_read_length << GEN6_SF_URB_ENTRY_READ_LENGTH_SHIFT |
+ urb_entry_read_offset << GEN6_SF_URB_ENTRY_READ_OFFSET_SHIFT;
+
+ dw2 = 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)
+ dw2 |= GEN6_SF_VIEWPORT_TRANSFORM_ENABLE;
+
dw3 = 0;
dw4 = 0;
dw16 = 0;
ctx->Point._Attenuated))
dw4 |= GEN6_SF_USE_STATE_POINT_WIDTH;
- dw4 |= U_FIXED(CLAMP(ctx->Point.Size, 0.125, 225.875), 3) <<
- GEN6_SF_POINT_WIDTH_SHIFT;
+ /* Clamp to ARB_point_parameters user limits */
+ point_size = CLAMP(ctx->Point.Size, ctx->Point.MinSize, ctx->Point.MaxSize);
+
+ /* Clamp to the hardware limits and convert to fixed point */
+ dw4 |= U_FIXED(CLAMP(point_size, 0.125, 255.875), 3);
+
if (ctx->Point.SpriteOrigin == GL_LOWER_LEFT)
dw1 |= GEN6_SF_POINT_SPRITE_LOWERLEFT;
(1 << GEN6_SF_TRIFAN_PROVOKE_SHIFT);
}
- if (ctx->Point.PointSprite) {
- for (i = 0; i < 8; i++) {
- if (ctx->Point.CoordReplace[i])
- dw16 |= (1 << i);
- }
- }
+ /* Create the mapping from the FS inputs we produce to the VS outputs
+ * they source from.
+ */
+ 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;
+
+ /* _NEW_POINT */
+ if (ctx->Point.PointSprite &&
+ (attr >= FRAG_ATTRIB_TEX0 && attr <= FRAG_ATTRIB_TEX7) &&
+ ctx->Point.CoordReplace[attr - FRAG_ATTRIB_TEX0]) {
+ dw16 |= (1 << input_index);
+ }
- /* flat shading */
- if (ctx->Light.ShadeModel == GL_FLAT) {
- dw17 |= ((brw->fragment_program->Base.InputsRead & (FRAG_BIT_COL0 | FRAG_BIT_COL1)) >>
- ((brw->fragment_program->Base.InputsRead & FRAG_BIT_WPOS) ? 0 : 1));
+ if (attr == FRAG_ATTRIB_PNTC)
+ dw16 |= (1 << input_index);
+
+ /* flat shading */
+ if (interp_qualifier == INTERP_QUALIFIER_FLAT ||
+ (shade_model_flat && is_gl_Color &&
+ interp_qualifier == INTERP_QUALIFIER_NONE))
+ dw17 |= (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);
}
+ for (; input_index < FRAG_ATTRIB_MAX; input_index++)
+ attr_overrides[input_index] = 0;
+
BEGIN_BATCH(20);
- OUT_BATCH(CMD_3D_SF_STATE << 16 | (20 - 2));
+ OUT_BATCH(_3DSTATE_SF << 16 | (20 - 2));
OUT_BATCH(dw1);
OUT_BATCH(dw2);
OUT_BATCH(dw3);
OUT_BATCH_F(ctx->Polygon.OffsetFactor); /* scale */
OUT_BATCH_F(0.0); /* XXX: global depth offset clamp */
for (i = 0; i < 8; i++) {
- uint32_t attr_overrides = 0;
-
- for (; attr < 64; attr++) {
- if (brw->fragment_program->Base.InputsRead & BITFIELD64_BIT(attr)) {
- attr_overrides |= get_attr_override(brw, attr, two_side_color);
- attr++;
- break;
- }
- }
-
- for (; attr < 64; attr++) {
- if (brw->fragment_program->Base.InputsRead & BITFIELD64_BIT(attr)) {
- attr_overrides |= get_attr_override(brw, attr, two_side_color) << 16;
- attr++;
- break;
- }
- }
- OUT_BATCH(attr_overrides);
+ OUT_BATCH(attr_overrides[i * 2] | attr_overrides[i * 2 + 1] << 16);
}
OUT_BATCH(dw16); /* point sprite texcoord bitmask */
OUT_BATCH(dw17); /* constant interp bitmask */
const struct brw_tracked_state gen6_sf_state = {
.dirty = {
.mesa = (_NEW_LIGHT |
+ _NEW_PROGRAM |
_NEW_POLYGON |
_NEW_LINE |
_NEW_SCISSOR |
_NEW_POINT |
_NEW_TRANSFORM),
.brw = (BRW_NEW_CONTEXT |
- BRW_NEW_FRAGMENT_PROGRAM),
+ BRW_NEW_FRAGMENT_PROGRAM |
+ BRW_NEW_HIZ),
.cache = CACHE_NEW_VS_PROG
},
.emit = upload_sf_state,
projects / mesa.git / blobdiff