1/*
2 * Copyright © 2009 Intel Corporation
3 *
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:
10 *
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
13 * Software.
14 *
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
21 * IN THE SOFTWARE.
22 *
23 * Authors:
24 *    Eric Anholt <eric@anholt.net>
25 *
26 */
27
28#include "brw_context.h"
29#include "brw_state.h"
30#include "brw_defines.h"
31#include "brw_util.h"
32#include "main/macros.h"
33#include "main/fbobject.h"
34#include "intel_batchbuffer.h"
35
36/**
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.
43 *
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.
48 *
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
53 * 256-bit increments.
54 */
55uint32_t
56get_attr_override(struct brw_vue_map *vue_map, int urb_entry_read_offset,
57                  int fs_attr, bool two_side_color, uint32_t *max_source_attr)
58{
59   int vs_attr = _mesa_frag_attrib_to_vert_result(fs_attr);
60   if (vs_attr < 0 || vs_attr == VERT_RESULT_HPOS) {
61      /* These attributes will be overwritten by the fragment shader's
62       * interpolation code (see emit_interp() in brw_wm_fp.c), so just let
63       * them reference the first available attribute.
64       */
65      return 0;
66   }
67
68   /* Find the VUE slot for this attribute. */
69   int slot = vue_map->vert_result_to_slot[vs_attr];
70
71   /* If there was only a back color written but not front, use back
72    * as the color instead of undefined
73    */
74   if (slot == -1 && vs_attr == VERT_RESULT_COL0)
75      slot = vue_map->vert_result_to_slot[VERT_RESULT_BFC0];
76   if (slot == -1 && vs_attr == VERT_RESULT_COL1)
77      slot = vue_map->vert_result_to_slot[VERT_RESULT_BFC1];
78
79   if (slot == -1) {
80      /* This attribute does not exist in the VUE--that means that the vertex
81       * shader did not write to it.  Behavior is undefined in this case, so
82       * just reference the first available attribute.
83       */
84      return 0;
85   }
86
87   /* Compute the location of the attribute relative to urb_entry_read_offset.
88    * Each increment of urb_entry_read_offset represents a 256-bit value, so
89    * it counts for two 128-bit VUE slots.
90    */
91   int source_attr = slot - 2 * urb_entry_read_offset;
92   assert(source_attr >= 0 && source_attr < 32);
93
94   /* If we are doing two-sided color, and the VUE slot following this one
95    * represents a back-facing color, then we need to instruct the SF unit to
96    * do back-facing swizzling.
97    */
98   bool swizzling = two_side_color &&
99      ((vue_map->slot_to_vert_result[slot] == VERT_RESULT_COL0 &&
100        vue_map->slot_to_vert_result[slot+1] == VERT_RESULT_BFC0) ||
101       (vue_map->slot_to_vert_result[slot] == VERT_RESULT_COL1 &&
102        vue_map->slot_to_vert_result[slot+1] == VERT_RESULT_BFC1));
103
104   /* Update max_source_attr.  If swizzling, the SF will read this slot + 1. */
105   if (*max_source_attr < source_attr + swizzling)
106      *max_source_attr = source_attr + swizzling;
107
108   if (swizzling) {
109      return source_attr |
110         (ATTRIBUTE_SWIZZLE_INPUTATTR_FACING << ATTRIBUTE_SWIZZLE_SHIFT);
111   }
112
113   return source_attr;
114}
115
116static void
117upload_sf_state(struct brw_context *brw)
118{
119   struct intel_context *intel = &brw->intel;
120   struct gl_context *ctx = &intel->ctx;
121   /* BRW_NEW_FRAGMENT_PROGRAM */
122   uint32_t num_outputs = _mesa_bitcount_64(brw->fragment_program->Base.InputsRead);
123   /* _NEW_LIGHT */
124   bool shade_model_flat = ctx->Light.ShadeModel == GL_FLAT;
125   uint32_t dw1, dw2, dw3, dw4, dw16, dw17;
126   int i;
127   /* _NEW_BUFFER */
128   bool render_to_fbo = _mesa_is_user_fbo(brw->intel.ctx.DrawBuffer);
129   bool multisampled_fbo = ctx->DrawBuffer->Visual.samples > 1;
130
131   int attr = 0, input_index = 0;
132   int urb_entry_read_offset = 1;
133   float point_size;
134   uint16_t attr_overrides[FRAG_ATTRIB_MAX];
135   uint32_t point_sprite_origin;
136
137   dw1 = GEN6_SF_SWIZZLE_ENABLE | num_outputs << GEN6_SF_NUM_OUTPUTS_SHIFT;
138
139   dw2 = GEN6_SF_STATISTICS_ENABLE |
140         GEN6_SF_VIEWPORT_TRANSFORM_ENABLE;
141
142   dw3 = 0;
143   dw4 = 0;
144   dw16 = 0;
145   dw17 = 0;
146
147   /* _NEW_POLYGON */
148   if ((ctx->Polygon.FrontFace == GL_CCW) ^ render_to_fbo)
149      dw2 |= GEN6_SF_WINDING_CCW;
150
151   if (ctx->Polygon.OffsetFill)
152       dw2 |= GEN6_SF_GLOBAL_DEPTH_OFFSET_SOLID;
153
154   if (ctx->Polygon.OffsetLine)
155       dw2 |= GEN6_SF_GLOBAL_DEPTH_OFFSET_WIREFRAME;
156
157   if (ctx->Polygon.OffsetPoint)
158       dw2 |= GEN6_SF_GLOBAL_DEPTH_OFFSET_POINT;
159
160   switch (ctx->Polygon.FrontMode) {
161   case GL_FILL:
162       dw2 |= GEN6_SF_FRONT_SOLID;
163       break;
164
165   case GL_LINE:
166       dw2 |= GEN6_SF_FRONT_WIREFRAME;
167       break;
168
169   case GL_POINT:
170       dw2 |= GEN6_SF_FRONT_POINT;
171       break;
172
173   default:
174       assert(0);
175       break;
176   }
177
178   switch (ctx->Polygon.BackMode) {
179   case GL_FILL:
180       dw2 |= GEN6_SF_BACK_SOLID;
181       break;
182
183   case GL_LINE:
184       dw2 |= GEN6_SF_BACK_WIREFRAME;
185       break;
186
187   case GL_POINT:
188       dw2 |= GEN6_SF_BACK_POINT;
189       break;
190
191   default:
192       assert(0);
193       break;
194   }
195
196   /* _NEW_SCISSOR */
197   if (ctx->Scissor.Enabled)
198      dw3 |= GEN6_SF_SCISSOR_ENABLE;
199
200   /* _NEW_POLYGON */
201   if (ctx->Polygon.CullFlag) {
202      switch (ctx->Polygon.CullFaceMode) {
203      case GL_FRONT:
204	 dw3 |= GEN6_SF_CULL_FRONT;
205	 break;
206      case GL_BACK:
207	 dw3 |= GEN6_SF_CULL_BACK;
208	 break;
209      case GL_FRONT_AND_BACK:
210	 dw3 |= GEN6_SF_CULL_BOTH;
211	 break;
212      default:
213	 assert(0);
214	 break;
215      }
216   } else {
217      dw3 |= GEN6_SF_CULL_NONE;
218   }
219
220   /* _NEW_LINE */
221   {
222      uint32_t line_width_u3_7 = U_FIXED(CLAMP(ctx->Line.Width, 0.0, 7.99), 7);
223      /* TODO: line width of 0 is not allowed when MSAA enabled */
224      if (line_width_u3_7 == 0)
225         line_width_u3_7 = 1;
226      dw3 |= line_width_u3_7 << GEN6_SF_LINE_WIDTH_SHIFT;
227   }
228   if (ctx->Line.SmoothFlag) {
229      dw3 |= GEN6_SF_LINE_AA_ENABLE;
230      dw3 |= GEN6_SF_LINE_AA_MODE_TRUE;
231      dw3 |= GEN6_SF_LINE_END_CAP_WIDTH_1_0;
232   }
233   /* _NEW_MULTISAMPLE */
234   if (multisampled_fbo && ctx->Multisample.Enabled)
235      dw3 |= GEN6_SF_MSRAST_ON_PATTERN;
236
237   /* _NEW_PROGRAM | _NEW_POINT */
238   if (!(ctx->VertexProgram.PointSizeEnabled ||
239	 ctx->Point._Attenuated))
240      dw4 |= GEN6_SF_USE_STATE_POINT_WIDTH;
241
242   /* Clamp to ARB_point_parameters user limits */
243   point_size = CLAMP(ctx->Point.Size, ctx->Point.MinSize, ctx->Point.MaxSize);
244
245   /* Clamp to the hardware limits and convert to fixed point */
246   dw4 |= U_FIXED(CLAMP(point_size, 0.125, 255.875), 3);
247
248   /*
249    * Window coordinates in an FBO are inverted, which means point
250    * sprite origin must be inverted, too.
251    */
252   if ((ctx->Point.SpriteOrigin == GL_LOWER_LEFT) != render_to_fbo) {
253      point_sprite_origin = GEN6_SF_POINT_SPRITE_LOWERLEFT;
254   } else {
255      point_sprite_origin = GEN6_SF_POINT_SPRITE_UPPERLEFT;
256   }
257   dw1 |= point_sprite_origin;
258
259   /* _NEW_LIGHT */
260   if (ctx->Light.ProvokingVertex != GL_FIRST_VERTEX_CONVENTION) {
261      dw4 |=
262	 (2 << GEN6_SF_TRI_PROVOKE_SHIFT) |
263	 (2 << GEN6_SF_TRIFAN_PROVOKE_SHIFT) |
264	 (1 << GEN6_SF_LINE_PROVOKE_SHIFT);
265   } else {
266      dw4 |=
267	 (1 << GEN6_SF_TRIFAN_PROVOKE_SHIFT);
268   }
269
270   /* Create the mapping from the FS inputs we produce to the VS outputs
271    * they source from.
272    */
273   uint32_t max_source_attr = 0;
274   for (; attr < FRAG_ATTRIB_MAX; attr++) {
275      enum glsl_interp_qualifier interp_qualifier =
276         brw->fragment_program->InterpQualifier[attr];
277      bool is_gl_Color = attr == FRAG_ATTRIB_COL0 || attr == FRAG_ATTRIB_COL1;
278
279      if (!(brw->fragment_program->Base.InputsRead & BITFIELD64_BIT(attr)))
280	 continue;
281
282      /* _NEW_POINT */
283      if (ctx->Point.PointSprite &&
284	  (attr >= FRAG_ATTRIB_TEX0 && attr <= FRAG_ATTRIB_TEX7) &&
285	  ctx->Point.CoordReplace[attr - FRAG_ATTRIB_TEX0]) {
286	 dw16 |= (1 << input_index);
287      }
288
289      if (attr == FRAG_ATTRIB_PNTC)
290	 dw16 |= (1 << input_index);
291
292      /* flat shading */
293      if (interp_qualifier == INTERP_QUALIFIER_FLAT ||
294          (shade_model_flat && is_gl_Color &&
295           interp_qualifier == INTERP_QUALIFIER_NONE))
296         dw17 |= (1 << input_index);
297
298      /* The hardware can only do the overrides on 16 overrides at a
299       * time, and the other up to 16 have to be lined up so that the
300       * input index = the output index.  We'll need to do some
301       * tweaking to make sure that's the case.
302       */
303      assert(input_index < 16 || attr == input_index);
304
305      /* CACHE_NEW_VS_PROG | _NEW_LIGHT | _NEW_PROGRAM */
306      attr_overrides[input_index++] =
307         get_attr_override(&brw->vs.prog_data->vue_map,
308			   urb_entry_read_offset, attr,
309                           ctx->VertexProgram._TwoSideEnabled,
310                           &max_source_attr);
311   }
312
313   for (; input_index < FRAG_ATTRIB_MAX; input_index++)
314      attr_overrides[input_index] = 0;
315
316   /* From the Sandy Bridge PRM, Volume 2, Part 1, documentation for
317    * 3DSTATE_SF DWord 1 bits 15:11, "Vertex URB Entry Read Length":
318    *
319    * "This field should be set to the minimum length required to read the
320    *  maximum source attribute.  The maximum source attribute is indicated
321    *  by the maximum value of the enabled Attribute # Source Attribute if
322    *  Attribute Swizzle Enable is set, Number of Output Attributes-1 if
323    *  enable is not set.
324    *  read_length = ceiling((max_source_attr + 1) / 2)
325    *
326    *  [errata] Corruption/Hang possible if length programmed larger than
327    *  recommended"
328    */
329   uint32_t urb_entry_read_length = ALIGN(max_source_attr + 1, 2) / 2;
330      dw1 |= urb_entry_read_length << GEN6_SF_URB_ENTRY_READ_LENGTH_SHIFT |
331             urb_entry_read_offset << GEN6_SF_URB_ENTRY_READ_OFFSET_SHIFT;
332
333   BEGIN_BATCH(20);
334   OUT_BATCH(_3DSTATE_SF << 16 | (20 - 2));
335   OUT_BATCH(dw1);
336   OUT_BATCH(dw2);
337   OUT_BATCH(dw3);
338   OUT_BATCH(dw4);
339   OUT_BATCH_F(ctx->Polygon.OffsetUnits * 2); /* constant.  copied from gen4 */
340   OUT_BATCH_F(ctx->Polygon.OffsetFactor); /* scale */
341   OUT_BATCH_F(0.0); /* XXX: global depth offset clamp */
342   for (i = 0; i < 8; i++) {
343      OUT_BATCH(attr_overrides[i * 2] | attr_overrides[i * 2 + 1] << 16);
344   }
345   OUT_BATCH(dw16); /* point sprite texcoord bitmask */
346   OUT_BATCH(dw17); /* constant interp bitmask */
347   OUT_BATCH(0); /* wrapshortest enables 0-7 */
348   OUT_BATCH(0); /* wrapshortest enables 8-15 */
349   ADVANCE_BATCH();
350}
351
352const struct brw_tracked_state gen6_sf_state = {
353   .dirty = {
354      .mesa  = (_NEW_LIGHT |
355		_NEW_PROGRAM |
356		_NEW_POLYGON |
357		_NEW_LINE |
358		_NEW_SCISSOR |
359		_NEW_BUFFERS |
360		_NEW_POINT |
361                _NEW_MULTISAMPLE),
362      .brw   = (BRW_NEW_CONTEXT |
363		BRW_NEW_FRAGMENT_PROGRAM),
364      .cache = CACHE_NEW_VS_PROG
365   },
366   .emit = upload_sf_state,
367};
368