2 * Copyright © 2009 Intel Corporation
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
24 * Eric Anholt <eric@anholt.net>
28 #include "brw_context.h"
29 #include "brw_state.h"
30 #include "brw_defines.h"
32 #include "main/macros.h"
33 #include "main/fbobject.h"
34 #include "intel_batchbuffer.h"
37 * Determine the appropriate attribute override value to store into the
38 * 3DSTATE_SF structure for a given fragment shader attribute. The attribute
39 * override value contains two pieces of information: the location of the
40 * attribute in the VUE (relative to urb_entry_read_offset, see below), and a
41 * flag indicating whether to "swizzle" the attribute based on the direction
42 * the triangle is facing.
44 * If an attribute is "swizzled", then the given VUE location is used for
45 * front-facing triangles, and the VUE location that immediately follows is
46 * used for back-facing triangles. We use this to implement the mapping from
47 * gl_FrontColor/gl_BackColor to gl_Color.
49 * urb_entry_read_offset is the offset into the VUE at which the SF unit is
50 * being instructed to begin reading attribute data. It can be set to a
51 * nonzero value to prevent the SF unit from wasting time reading elements of
52 * the VUE that are not needed by the fragment shader. It is measured in
56 get_attr_override(struct brw_vue_map *vue_map, int urb_entry_read_offset,
57 int fs_attr, bool two_side_color)
59 int attr_override, slot;
60 int vs_attr = _mesa_frag_attrib_to_vert_result(fs_attr);
61 if (vs_attr < 0 || vs_attr == VERT_RESULT_HPOS) {
62 /* These attributes will be overwritten by the fragment shader's
63 * interpolation code (see emit_interp() in brw_wm_fp.c), so just let
64 * them reference the first available attribute.
69 /* Find the VUE slot for this attribute. */
70 slot = vue_map->vert_result_to_slot[vs_attr];
72 /* If there was only a back color written but not front, use back
73 * as the color instead of undefined
75 if (slot == -1 && vs_attr == VERT_RESULT_COL0)
76 slot = vue_map->vert_result_to_slot[VERT_RESULT_BFC0];
77 if (slot == -1 && vs_attr == VERT_RESULT_COL1)
78 slot = vue_map->vert_result_to_slot[VERT_RESULT_BFC1];
81 /* This attribute does not exist in the VUE--that means that the vertex
82 * shader did not write to it. Behavior is undefined in this case, so
83 * just reference the first available attribute.
88 /* Compute the location of the attribute relative to urb_entry_read_offset.
89 * Each increment of urb_entry_read_offset represents a 256-bit value, so
90 * it counts for two 128-bit VUE slots.
92 attr_override = slot - 2 * urb_entry_read_offset;
93 assert (attr_override >= 0 && attr_override < 32);
95 /* If we are doing two-sided color, and the VUE slot following this one
96 * represents a back-facing color, then we need to instruct the SF unit to
97 * do back-facing swizzling.
100 if (vue_map->slot_to_vert_result[slot] == VERT_RESULT_COL0 &&
101 vue_map->slot_to_vert_result[slot+1] == VERT_RESULT_BFC0)
102 attr_override |= (ATTRIBUTE_SWIZZLE_INPUTATTR_FACING << ATTRIBUTE_SWIZZLE_SHIFT);
103 else if (vue_map->slot_to_vert_result[slot] == VERT_RESULT_COL1 &&
104 vue_map->slot_to_vert_result[slot+1] == VERT_RESULT_BFC1)
105 attr_override |= (ATTRIBUTE_SWIZZLE_INPUTATTR_FACING << ATTRIBUTE_SWIZZLE_SHIFT);
108 return attr_override;
112 upload_sf_state(struct brw_context *brw)
114 struct intel_context *intel = &brw->intel;
115 struct gl_context *ctx = &intel->ctx;
116 uint32_t urb_entry_read_length;
117 /* BRW_NEW_FRAGMENT_PROGRAM */
118 uint32_t num_outputs = _mesa_bitcount_64(brw->fragment_program->Base.InputsRead);
120 bool shade_model_flat = ctx->Light.ShadeModel == GL_FLAT;
121 uint32_t dw1, dw2, dw3, dw4, dw16, dw17;
124 bool render_to_fbo = _mesa_is_user_fbo(brw->intel.ctx.DrawBuffer);
125 bool multisampled = false;
126 if (ctx->DrawBuffer->_ColorDrawBuffers[0])
127 multisampled = ctx->DrawBuffer->_ColorDrawBuffers[0]->NumSamples > 0;
129 int attr = 0, input_index = 0;
130 int urb_entry_read_offset = 1;
132 uint16_t attr_overrides[FRAG_ATTRIB_MAX];
133 uint32_t point_sprite_origin;
135 /* CACHE_NEW_VS_PROG */
136 urb_entry_read_length = ((brw->vs.prog_data->vue_map.num_slots + 1) / 2 -
137 urb_entry_read_offset);
138 if (urb_entry_read_length == 0) {
139 /* Setting the URB entry read length to 0 causes undefined behavior, so
140 * if we have no URB data to read, set it to 1.
142 urb_entry_read_length = 1;
146 GEN6_SF_SWIZZLE_ENABLE |
147 num_outputs << GEN6_SF_NUM_OUTPUTS_SHIFT |
148 urb_entry_read_length << GEN6_SF_URB_ENTRY_READ_LENGTH_SHIFT |
149 urb_entry_read_offset << GEN6_SF_URB_ENTRY_READ_OFFSET_SHIFT;
151 dw2 = GEN6_SF_STATISTICS_ENABLE |
152 GEN6_SF_VIEWPORT_TRANSFORM_ENABLE;
160 if ((ctx->Polygon.FrontFace == GL_CCW) ^ render_to_fbo)
161 dw2 |= GEN6_SF_WINDING_CCW;
163 if (ctx->Polygon.OffsetFill)
164 dw2 |= GEN6_SF_GLOBAL_DEPTH_OFFSET_SOLID;
166 if (ctx->Polygon.OffsetLine)
167 dw2 |= GEN6_SF_GLOBAL_DEPTH_OFFSET_WIREFRAME;
169 if (ctx->Polygon.OffsetPoint)
170 dw2 |= GEN6_SF_GLOBAL_DEPTH_OFFSET_POINT;
172 switch (ctx->Polygon.FrontMode) {
174 dw2 |= GEN6_SF_FRONT_SOLID;
178 dw2 |= GEN6_SF_FRONT_WIREFRAME;
182 dw2 |= GEN6_SF_FRONT_POINT;
190 switch (ctx->Polygon.BackMode) {
192 dw2 |= GEN6_SF_BACK_SOLID;
196 dw2 |= GEN6_SF_BACK_WIREFRAME;
200 dw2 |= GEN6_SF_BACK_POINT;
209 if (ctx->Scissor.Enabled)
210 dw3 |= GEN6_SF_SCISSOR_ENABLE;
213 if (ctx->Polygon.CullFlag) {
214 switch (ctx->Polygon.CullFaceMode) {
216 dw3 |= GEN6_SF_CULL_FRONT;
219 dw3 |= GEN6_SF_CULL_BACK;
221 case GL_FRONT_AND_BACK:
222 dw3 |= GEN6_SF_CULL_BOTH;
229 dw3 |= GEN6_SF_CULL_NONE;
234 uint32_t line_width_u3_7 = U_FIXED(CLAMP(ctx->Line.Width, 0.0, 7.99), 7);
235 /* TODO: line width of 0 is not allowed when MSAA enabled */
236 if (line_width_u3_7 == 0)
238 dw3 |= line_width_u3_7 << GEN6_SF_LINE_WIDTH_SHIFT;
240 if (ctx->Line.SmoothFlag) {
241 dw3 |= GEN6_SF_LINE_AA_ENABLE;
242 dw3 |= GEN6_SF_LINE_AA_MODE_TRUE;
243 dw3 |= GEN6_SF_LINE_END_CAP_WIDTH_1_0;
246 dw3 |= GEN6_SF_MSRAST_ON_PATTERN;
248 /* _NEW_PROGRAM | _NEW_POINT */
249 if (!(ctx->VertexProgram.PointSizeEnabled ||
250 ctx->Point._Attenuated))
251 dw4 |= GEN6_SF_USE_STATE_POINT_WIDTH;
253 /* Clamp to ARB_point_parameters user limits */
254 point_size = CLAMP(ctx->Point.Size, ctx->Point.MinSize, ctx->Point.MaxSize);
256 /* Clamp to the hardware limits and convert to fixed point */
257 dw4 |= U_FIXED(CLAMP(point_size, 0.125, 255.875), 3);
260 * Window coordinates in an FBO are inverted, which means point
261 * sprite origin must be inverted, too.
263 if ((ctx->Point.SpriteOrigin == GL_LOWER_LEFT) != render_to_fbo) {
264 point_sprite_origin = GEN6_SF_POINT_SPRITE_LOWERLEFT;
266 point_sprite_origin = GEN6_SF_POINT_SPRITE_UPPERLEFT;
268 dw1 |= point_sprite_origin;
271 if (ctx->Light.ProvokingVertex != GL_FIRST_VERTEX_CONVENTION) {
273 (2 << GEN6_SF_TRI_PROVOKE_SHIFT) |
274 (2 << GEN6_SF_TRIFAN_PROVOKE_SHIFT) |
275 (1 << GEN6_SF_LINE_PROVOKE_SHIFT);
278 (1 << GEN6_SF_TRIFAN_PROVOKE_SHIFT);
281 /* Create the mapping from the FS inputs we produce to the VS outputs
284 for (; attr < FRAG_ATTRIB_MAX; attr++) {
285 enum glsl_interp_qualifier interp_qualifier =
286 brw->fragment_program->InterpQualifier[attr];
287 bool is_gl_Color = attr == FRAG_ATTRIB_COL0 || attr == FRAG_ATTRIB_COL1;
289 if (!(brw->fragment_program->Base.InputsRead & BITFIELD64_BIT(attr)))
293 if (ctx->Point.PointSprite &&
294 (attr >= FRAG_ATTRIB_TEX0 && attr <= FRAG_ATTRIB_TEX7) &&
295 ctx->Point.CoordReplace[attr - FRAG_ATTRIB_TEX0]) {
296 dw16 |= (1 << input_index);
299 if (attr == FRAG_ATTRIB_PNTC)
300 dw16 |= (1 << input_index);
303 if (interp_qualifier == INTERP_QUALIFIER_FLAT ||
304 (shade_model_flat && is_gl_Color &&
305 interp_qualifier == INTERP_QUALIFIER_NONE))
306 dw17 |= (1 << input_index);
308 /* The hardware can only do the overrides on 16 overrides at a
309 * time, and the other up to 16 have to be lined up so that the
310 * input index = the output index. We'll need to do some
311 * tweaking to make sure that's the case.
313 assert(input_index < 16 || attr == input_index);
315 /* CACHE_NEW_VS_PROG | _NEW_LIGHT | _NEW_PROGRAM */
316 attr_overrides[input_index++] =
317 get_attr_override(&brw->vs.prog_data->vue_map,
318 urb_entry_read_offset, attr,
319 ctx->VertexProgram._TwoSideEnabled);
322 for (; input_index < FRAG_ATTRIB_MAX; input_index++)
323 attr_overrides[input_index] = 0;
326 OUT_BATCH(_3DSTATE_SF << 16 | (20 - 2));
331 OUT_BATCH_F(ctx->Polygon.OffsetUnits * 2); /* constant. copied from gen4 */
332 OUT_BATCH_F(ctx->Polygon.OffsetFactor); /* scale */
333 OUT_BATCH_F(0.0); /* XXX: global depth offset clamp */
334 for (i = 0; i < 8; i++) {
335 OUT_BATCH(attr_overrides[i * 2] | attr_overrides[i * 2 + 1] << 16);
337 OUT_BATCH(dw16); /* point sprite texcoord bitmask */
338 OUT_BATCH(dw17); /* constant interp bitmask */
339 OUT_BATCH(0); /* wrapshortest enables 0-7 */
340 OUT_BATCH(0); /* wrapshortest enables 8-15 */
344 const struct brw_tracked_state gen6_sf_state = {
346 .mesa = (_NEW_LIGHT |
353 .brw = (BRW_NEW_CONTEXT |
354 BRW_NEW_FRAGMENT_PROGRAM),
355 .cache = CACHE_NEW_VS_PROG
357 .emit = upload_sf_state,