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 "compiler/nir/nir.h"
33#include "main/macros.h"
34#include "main/fbobject.h"
35#include "main/framebuffer.h"
36#include "intel_batchbuffer.h"
37
38/**
39 * Determine the appropriate attribute override value to store into the
40 * 3DSTATE_SF structure for a given fragment shader attribute.  The attribute
41 * override value contains two pieces of information: the location of the
42 * attribute in the VUE (relative to urb_entry_read_offset, see below), and a
43 * flag indicating whether to "swizzle" the attribute based on the direction
44 * the triangle is facing.
45 *
46 * If an attribute is "swizzled", then the given VUE location is used for
47 * front-facing triangles, and the VUE location that immediately follows is
48 * used for back-facing triangles.  We use this to implement the mapping from
49 * gl_FrontColor/gl_BackColor to gl_Color.
50 *
51 * urb_entry_read_offset is the offset into the VUE at which the SF unit is
52 * being instructed to begin reading attribute data.  It can be set to a
53 * nonzero value to prevent the SF unit from wasting time reading elements of
54 * the VUE that are not needed by the fragment shader.  It is measured in
55 * 256-bit increments.
56 */
57static uint32_t
58get_attr_override(const struct brw_vue_map *vue_map, int urb_entry_read_offset,
59                  int fs_attr, bool two_side_color, uint32_t *max_source_attr)
60{
61   /* Find the VUE slot for this attribute. */
62   int slot = vue_map->varying_to_slot[fs_attr];
63
64   /* Viewport and Layer are stored in the VUE header.  We need to override
65    * them to zero if earlier stages didn't write them, as GL requires that
66    * they read back as zero when not explicitly set.
67    */
68   if (fs_attr == VARYING_SLOT_VIEWPORT || fs_attr == VARYING_SLOT_LAYER) {
69      unsigned override =
70         ATTRIBUTE_0_OVERRIDE_X | ATTRIBUTE_0_OVERRIDE_W |
71         ATTRIBUTE_CONST_0000 << ATTRIBUTE_0_CONST_SOURCE_SHIFT;
72
73      if (!(vue_map->slots_valid & VARYING_BIT_LAYER))
74         override |= ATTRIBUTE_0_OVERRIDE_Y;
75      if (!(vue_map->slots_valid & VARYING_BIT_VIEWPORT))
76         override |= ATTRIBUTE_0_OVERRIDE_Z;
77
78      return override;
79   }
80
81   /* If there was only a back color written but not front, use back
82    * as the color instead of undefined
83    */
84   if (slot == -1 && fs_attr == VARYING_SLOT_COL0)
85      slot = vue_map->varying_to_slot[VARYING_SLOT_BFC0];
86   if (slot == -1 && fs_attr == VARYING_SLOT_COL1)
87      slot = vue_map->varying_to_slot[VARYING_SLOT_BFC1];
88
89   if (slot == -1) {
90      /* This attribute does not exist in the VUE--that means that the vertex
91       * shader did not write to it.  This means that either:
92       *
93       * (a) This attribute is a texture coordinate, and it is going to be
94       * replaced with point coordinates (as a consequence of a call to
95       * glTexEnvi(GL_POINT_SPRITE, GL_COORD_REPLACE, GL_TRUE)), so the
96       * hardware will ignore whatever attribute override we supply.
97       *
98       * (b) This attribute is read by the fragment shader but not written by
99       * the vertex shader, so its value is undefined.  Therefore the
100       * attribute override we supply doesn't matter.
101       *
102       * (c) This attribute is gl_PrimitiveID, and it wasn't written by the
103       * previous shader stage.
104       *
105       * Note that we don't have to worry about the cases where the attribute
106       * is gl_PointCoord or is undergoing point sprite coordinate
107       * replacement, because in those cases, this function isn't called.
108       *
109       * In case (c), we need to program the attribute overrides so that the
110       * primitive ID will be stored in this slot.  In every other case, the
111       * attribute override we supply doesn't matter.  So just go ahead and
112       * program primitive ID in every case.
113       */
114      return (ATTRIBUTE_0_OVERRIDE_W |
115              ATTRIBUTE_0_OVERRIDE_Z |
116              ATTRIBUTE_0_OVERRIDE_Y |
117              ATTRIBUTE_0_OVERRIDE_X |
118              (ATTRIBUTE_CONST_PRIM_ID << ATTRIBUTE_0_CONST_SOURCE_SHIFT));
119   }
120
121   /* Compute the location of the attribute relative to urb_entry_read_offset.
122    * Each increment of urb_entry_read_offset represents a 256-bit value, so
123    * it counts for two 128-bit VUE slots.
124    */
125   int source_attr = slot - 2 * urb_entry_read_offset;
126   assert(source_attr >= 0 && source_attr < 32);
127
128   /* If we are doing two-sided color, and the VUE slot following this one
129    * represents a back-facing color, then we need to instruct the SF unit to
130    * do back-facing swizzling.
131    */
132   bool swizzling = two_side_color &&
133      ((vue_map->slot_to_varying[slot] == VARYING_SLOT_COL0 &&
134        vue_map->slot_to_varying[slot+1] == VARYING_SLOT_BFC0) ||
135       (vue_map->slot_to_varying[slot] == VARYING_SLOT_COL1 &&
136        vue_map->slot_to_varying[slot+1] == VARYING_SLOT_BFC1));
137
138   /* Update max_source_attr.  If swizzling, the SF will read this slot + 1. */
139   if (*max_source_attr < source_attr + swizzling)
140      *max_source_attr = source_attr + swizzling;
141
142   if (swizzling) {
143      return source_attr |
144         (ATTRIBUTE_SWIZZLE_INPUTATTR_FACING << ATTRIBUTE_SWIZZLE_SHIFT);
145   }
146
147   return source_attr;
148}
149
150
151/**
152 * Create the mapping from the FS inputs we produce to the previous pipeline
153 * stage (GS or VS) outputs they source from.
154 */
155void
156calculate_attr_overrides(const struct brw_context *brw,
157                         uint16_t *attr_overrides,
158                         uint32_t *point_sprite_enables,
159                         uint32_t *urb_entry_read_length,
160                         uint32_t *urb_entry_read_offset)
161{
162   /* BRW_NEW_FS_PROG_DATA */
163   const struct brw_wm_prog_data *wm_prog_data =
164      brw_wm_prog_data(brw->wm.base.prog_data);
165   uint32_t max_source_attr = 0;
166
167   *point_sprite_enables = 0;
168
169   *urb_entry_read_offset = BRW_SF_URB_ENTRY_READ_OFFSET;
170
171   /* BRW_NEW_FRAGMENT_PROGRAM
172    *
173    * If the fragment shader reads VARYING_SLOT_LAYER, then we need to pass in
174    * the full vertex header.  Otherwise, we can program the SF to start
175    * reading at an offset of 1 (2 varying slots) to skip unnecessary data:
176    * - VARYING_SLOT_PSIZ and BRW_VARYING_SLOT_NDC on gen4-5
177    * - VARYING_SLOT_{PSIZ,LAYER} and VARYING_SLOT_POS on gen6+
178    */
179
180   bool fs_needs_vue_header = brw->fragment_program->info.inputs_read &
181      (VARYING_BIT_LAYER | VARYING_BIT_VIEWPORT);
182
183   *urb_entry_read_offset = fs_needs_vue_header ? 0 : 1;
184
185   /* From the Ivybridge PRM, Vol 2 Part 1, 3DSTATE_SBE,
186    * description of dw10 Point Sprite Texture Coordinate Enable:
187    *
188    * "This field must be programmed to zero when non-point primitives
189    * are rendered."
190    *
191    * The SandyBridge PRM doesn't explicitly say that point sprite enables
192    * must be programmed to zero when rendering non-point primitives, but
193    * the IvyBridge PRM does, and if we don't, we get garbage.
194    *
195    * This is not required on Haswell, as the hardware ignores this state
196    * when drawing non-points -- although we do still need to be careful to
197    * correctly set the attr overrides.
198    *
199    * _NEW_POLYGON
200    * BRW_NEW_PRIMITIVE | BRW_NEW_GS_PROG_DATA | BRW_NEW_TES_PROG_DATA
201    */
202   bool drawing_points = brw_is_drawing_points(brw);
203
204   /* Initialize all the attr_overrides to 0.  In the loop below we'll modify
205    * just the ones that correspond to inputs used by the fs.
206    */
207   memset(attr_overrides, 0, 16*sizeof(*attr_overrides));
208
209   for (int attr = 0; attr < VARYING_SLOT_MAX; attr++) {
210      int input_index = wm_prog_data->urb_setup[attr];
211
212      if (input_index < 0)
213	 continue;
214
215      /* _NEW_POINT */
216      bool point_sprite = false;
217      if (drawing_points) {
218         if (brw->ctx.Point.PointSprite &&
219             (attr >= VARYING_SLOT_TEX0 && attr <= VARYING_SLOT_TEX7) &&
220             (brw->ctx.Point.CoordReplace & (1u << (attr - VARYING_SLOT_TEX0)))) {
221            point_sprite = true;
222         }
223
224         if (attr == VARYING_SLOT_PNTC)
225            point_sprite = true;
226
227         if (point_sprite)
228            *point_sprite_enables |= (1 << input_index);
229      }
230
231      /* BRW_NEW_VUE_MAP_GEOM_OUT | _NEW_LIGHT | _NEW_PROGRAM */
232      uint16_t attr_override = point_sprite ? 0 :
233         get_attr_override(&brw->vue_map_geom_out,
234			   *urb_entry_read_offset, attr,
235                           brw->ctx.VertexProgram._TwoSideEnabled,
236                           &max_source_attr);
237
238      /* The hardware can only do the overrides on 16 overrides at a
239       * time, and the other up to 16 have to be lined up so that the
240       * input index = the output index.  We'll need to do some
241       * tweaking to make sure that's the case.
242       */
243      if (input_index < 16)
244         attr_overrides[input_index] = attr_override;
245      else
246         assert(attr_override == input_index);
247   }
248
249   /* From the Sandy Bridge PRM, Volume 2, Part 1, documentation for
250    * 3DSTATE_SF DWord 1 bits 15:11, "Vertex URB Entry Read Length":
251    *
252    * "This field should be set to the minimum length required to read the
253    *  maximum source attribute.  The maximum source attribute is indicated
254    *  by the maximum value of the enabled Attribute # Source Attribute if
255    *  Attribute Swizzle Enable is set, Number of Output Attributes-1 if
256    *  enable is not set.
257    *  read_length = ceiling((max_source_attr + 1) / 2)
258    *
259    *  [errata] Corruption/Hang possible if length programmed larger than
260    *  recommended"
261    *
262    * Similar text exists for Ivy Bridge.
263    */
264   *urb_entry_read_length = ALIGN(max_source_attr + 1, 2) / 2;
265}
266
267
268static void
269upload_sf_state(struct brw_context *brw)
270{
271   struct gl_context *ctx = &brw->ctx;
272   /* BRW_NEW_FS_PROG_DATA */
273   const struct brw_wm_prog_data *wm_prog_data =
274      brw_wm_prog_data(brw->wm.base.prog_data);
275   uint32_t num_outputs = wm_prog_data->num_varying_inputs;
276   uint32_t dw1, dw2, dw3, dw4;
277   uint32_t point_sprite_enables;
278   int i;
279   /* _NEW_BUFFER */
280   bool render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer);
281   const bool multisampled_fbo = _mesa_geometric_samples(ctx->DrawBuffer) > 1;
282
283   float point_size;
284   uint16_t attr_overrides[16];
285   uint32_t point_sprite_origin;
286
287   dw1 = GEN6_SF_SWIZZLE_ENABLE | num_outputs << GEN6_SF_NUM_OUTPUTS_SHIFT;
288   dw2 = GEN6_SF_STATISTICS_ENABLE;
289   dw3 = GEN6_SF_SCISSOR_ENABLE;
290   dw4 = 0;
291
292   if (brw->sf.viewport_transform_enable)
293       dw2 |= GEN6_SF_VIEWPORT_TRANSFORM_ENABLE;
294
295   /* _NEW_POLYGON */
296   if (ctx->Polygon._FrontBit == render_to_fbo)
297      dw2 |= GEN6_SF_WINDING_CCW;
298
299   if (ctx->Polygon.OffsetFill)
300       dw2 |= GEN6_SF_GLOBAL_DEPTH_OFFSET_SOLID;
301
302   if (ctx->Polygon.OffsetLine)
303       dw2 |= GEN6_SF_GLOBAL_DEPTH_OFFSET_WIREFRAME;
304
305   if (ctx->Polygon.OffsetPoint)
306       dw2 |= GEN6_SF_GLOBAL_DEPTH_OFFSET_POINT;
307
308   switch (ctx->Polygon.FrontMode) {
309   case GL_FILL:
310       dw2 |= GEN6_SF_FRONT_SOLID;
311       break;
312
313   case GL_LINE:
314       dw2 |= GEN6_SF_FRONT_WIREFRAME;
315       break;
316
317   case GL_POINT:
318       dw2 |= GEN6_SF_FRONT_POINT;
319       break;
320
321   default:
322       unreachable("not reached");
323   }
324
325   switch (ctx->Polygon.BackMode) {
326   case GL_FILL:
327       dw2 |= GEN6_SF_BACK_SOLID;
328       break;
329
330   case GL_LINE:
331       dw2 |= GEN6_SF_BACK_WIREFRAME;
332       break;
333
334   case GL_POINT:
335       dw2 |= GEN6_SF_BACK_POINT;
336       break;
337
338   default:
339       unreachable("not reached");
340   }
341
342   /* _NEW_POLYGON */
343   if (ctx->Polygon.CullFlag) {
344      switch (ctx->Polygon.CullFaceMode) {
345      case GL_FRONT:
346	 dw3 |= GEN6_SF_CULL_FRONT;
347	 break;
348      case GL_BACK:
349	 dw3 |= GEN6_SF_CULL_BACK;
350	 break;
351      case GL_FRONT_AND_BACK:
352	 dw3 |= GEN6_SF_CULL_BOTH;
353	 break;
354      default:
355	 unreachable("not reached");
356      }
357   } else {
358      dw3 |= GEN6_SF_CULL_NONE;
359   }
360
361   /* _NEW_LINE */
362   {
363      uint32_t line_width_u3_7 = brw_get_line_width(brw);
364      dw3 |= line_width_u3_7 << GEN6_SF_LINE_WIDTH_SHIFT;
365   }
366   if (ctx->Line.SmoothFlag) {
367      dw3 |= GEN6_SF_LINE_AA_ENABLE;
368      dw3 |= GEN6_SF_LINE_AA_MODE_TRUE;
369      dw3 |= GEN6_SF_LINE_END_CAP_WIDTH_1_0;
370   }
371   /* _NEW_MULTISAMPLE */
372   if (multisampled_fbo && ctx->Multisample.Enabled)
373      dw3 |= GEN6_SF_MSRAST_ON_PATTERN;
374
375   /* _NEW_PROGRAM | _NEW_POINT, BRW_NEW_VUE_MAP_GEOM_OUT */
376   if (use_state_point_size(brw))
377      dw4 |= GEN6_SF_USE_STATE_POINT_WIDTH;
378
379   /* _NEW_POINT - Clamp to ARB_point_parameters user limits */
380   point_size = CLAMP(ctx->Point.Size, ctx->Point.MinSize, ctx->Point.MaxSize);
381
382   /* Clamp to the hardware limits and convert to fixed point */
383   dw4 |= U_FIXED(CLAMP(point_size, 0.125f, 255.875f), 3);
384
385   /*
386    * Window coordinates in an FBO are inverted, which means point
387    * sprite origin must be inverted, too.
388    */
389   if ((ctx->Point.SpriteOrigin == GL_LOWER_LEFT) != render_to_fbo) {
390      point_sprite_origin = GEN6_SF_POINT_SPRITE_LOWERLEFT;
391   } else {
392      point_sprite_origin = GEN6_SF_POINT_SPRITE_UPPERLEFT;
393   }
394   dw1 |= point_sprite_origin;
395
396   /* _NEW_LIGHT */
397   if (ctx->Light.ProvokingVertex != GL_FIRST_VERTEX_CONVENTION) {
398      dw4 |=
399	 (2 << GEN6_SF_TRI_PROVOKE_SHIFT) |
400	 (2 << GEN6_SF_TRIFAN_PROVOKE_SHIFT) |
401	 (1 << GEN6_SF_LINE_PROVOKE_SHIFT);
402   } else {
403      dw4 |=
404	 (1 << GEN6_SF_TRIFAN_PROVOKE_SHIFT);
405   }
406
407   /* BRW_NEW_VUE_MAP_GEOM_OUT | BRW_NEW_FRAGMENT_PROGRAM |
408    * _NEW_POINT | _NEW_LIGHT | _NEW_PROGRAM | BRW_NEW_FS_PROG_DATA
409    */
410   uint32_t urb_entry_read_length;
411   uint32_t urb_entry_read_offset;
412   calculate_attr_overrides(brw, attr_overrides, &point_sprite_enables,
413                            &urb_entry_read_length, &urb_entry_read_offset);
414   dw1 |= (urb_entry_read_length << GEN6_SF_URB_ENTRY_READ_LENGTH_SHIFT |
415           urb_entry_read_offset << GEN6_SF_URB_ENTRY_READ_OFFSET_SHIFT);
416
417   BEGIN_BATCH(20);
418   OUT_BATCH(_3DSTATE_SF << 16 | (20 - 2));
419   OUT_BATCH(dw1);
420   OUT_BATCH(dw2);
421   OUT_BATCH(dw3);
422   OUT_BATCH(dw4);
423   OUT_BATCH_F(ctx->Polygon.OffsetUnits * 2); /* constant.  copied from gen4 */
424   OUT_BATCH_F(ctx->Polygon.OffsetFactor); /* scale */
425   OUT_BATCH_F(ctx->Polygon.OffsetClamp); /* global depth offset clamp */
426   for (i = 0; i < 8; i++) {
427      OUT_BATCH(attr_overrides[i * 2] | attr_overrides[i * 2 + 1] << 16);
428   }
429   OUT_BATCH(point_sprite_enables); /* dw16 */
430   OUT_BATCH(wm_prog_data->flat_inputs);
431   OUT_BATCH(0); /* wrapshortest enables 0-7 */
432   OUT_BATCH(0); /* wrapshortest enables 8-15 */
433   ADVANCE_BATCH();
434}
435
436const struct brw_tracked_state gen6_sf_state = {
437   .dirty = {
438      .mesa  = _NEW_BUFFERS |
439               _NEW_LIGHT |
440               _NEW_LINE |
441               _NEW_MULTISAMPLE |
442               _NEW_POINT |
443               _NEW_POLYGON |
444               _NEW_PROGRAM,
445      .brw   = BRW_NEW_BLORP |
446               BRW_NEW_CONTEXT |
447               BRW_NEW_FRAGMENT_PROGRAM |
448               BRW_NEW_FS_PROG_DATA |
449               BRW_NEW_GS_PROG_DATA |
450               BRW_NEW_PRIMITIVE |
451               BRW_NEW_TES_PROG_DATA |
452               BRW_NEW_VUE_MAP_GEOM_OUT,
453   },
454   .emit = upload_sf_state,
455};
456