1/*
2 * Copyright © 2019 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#ifndef INTEL_GUARDBAND_H
24#define INTEL_GUARDBAND_H
25
26static inline void
27intel_calculate_guardband_size(uint32_t x_min, uint32_t x_max,
28                               uint32_t y_min, uint32_t y_max,
29                               float m00, float m11, float m30, float m31,
30                               float *xmin, float *xmax,
31                               float *ymin, float *ymax)
32{
33   /* According to the "Vertex X,Y Clamping and Quantization" section of the
34    * Strips and Fans documentation:
35    *
36    * "The vertex X and Y screen-space coordinates are also /clamped/ to the
37    *  fixed-point "guardband" range supported by the rasterization hardware"
38    *
39    * and
40    *
41    * "In almost all circumstances, if an object’s vertices are actually
42    *  modified by this clamping (i.e., had X or Y coordinates outside of
43    *  the guardband extent the rendered object will not match the intended
44    *  result.  Therefore software should take steps to ensure that this does
45    *  not happen - e.g., by clipping objects such that they do not exceed
46    *  these limits after the Drawing Rectangle is applied."
47    *
48    * I believe the fundamental restriction is that the rasterizer (in
49    * the SF/WM stages) have a limit on the number of pixels that can be
50    * rasterized.  We need to ensure any coordinates beyond the rasterizer
51    * limit are handled by the clipper.  So effectively that limit becomes
52    * the clipper's guardband size.
53    *
54    * It goes on to say:
55    *
56    * "In addition, in order to be correctly rendered, objects must have a
57    *  screenspace bounding box not exceeding 8K in the X or Y direction.
58    *  This additional restriction must also be comprehended by software,
59    *  i.e., enforced by use of clipping."
60    *
61    * This makes no sense.  Gfx7+ hardware supports 16K render targets,
62    * and you definitely need to be able to draw polygons that fill the
63    * surface.  Our assumption is that the rasterizer was limited to 8K
64    * on Sandybridge, which only supports 8K surfaces, and it was actually
65    * increased to 16K on Ivybridge and later.
66    *
67    * So, limit the guardband to 16K on Gfx7+ and 8K on Sandybridge.
68    */
69   const float gb_size = GFX_VER >= 7 ? 16384.0f : 8192.0f;
70
71   /* Workaround: prevent gpu hangs on SandyBridge
72    * by disabling guardband clipping for odd dimensions.
73    */
74   if (GFX_VER == 6 && (x_min & 1 || x_max & 1 || y_min & 1 || y_max & 1)) {
75      *xmin = -1.0f;
76      *xmax =  1.0f;
77      *ymin = -1.0f;
78      *ymax =  1.0f;
79      return;
80   }
81
82   if (m00 != 0 && m11 != 0) {
83      /* First, we compute the screen-space render area */
84      const float ss_ra_xmin = MIN3(x_min, m30 + m00, m30 - m00);
85      const float ss_ra_xmax = MAX3(x_max, m30 + m00, m30 - m00);
86      const float ss_ra_ymin = MIN3(y_min, m31 + m11, m31 - m11);
87      const float ss_ra_ymax = MAX3(y_max, m31 + m11, m31 - m11);
88
89      /* We want the guardband to be centered on that */
90      const float ss_gb_xmin = (ss_ra_xmin + ss_ra_xmax) / 2 - gb_size;
91      const float ss_gb_xmax = (ss_ra_xmin + ss_ra_xmax) / 2 + gb_size;
92      const float ss_gb_ymin = (ss_ra_ymin + ss_ra_ymax) / 2 - gb_size;
93      const float ss_gb_ymax = (ss_ra_ymin + ss_ra_ymax) / 2 + gb_size;
94
95      /* Now we need it in native device coordinates */
96      const float ndc_gb_xmin = (ss_gb_xmin - m30) / m00;
97      const float ndc_gb_xmax = (ss_gb_xmax - m30) / m00;
98      const float ndc_gb_ymin = (ss_gb_ymin - m31) / m11;
99      const float ndc_gb_ymax = (ss_gb_ymax - m31) / m11;
100
101      /* Thanks to Y-flipping and ORIGIN_UPPER_LEFT, the Y coordinates may be
102       * flipped upside-down.  X should be fine though.
103       */
104      assert(ndc_gb_xmin <= ndc_gb_xmax);
105      *xmin = ndc_gb_xmin;
106      *xmax = ndc_gb_xmax;
107      *ymin = MIN2(ndc_gb_ymin, ndc_gb_ymax);
108      *ymax = MAX2(ndc_gb_ymin, ndc_gb_ymax);
109   } else {
110      /* The viewport scales to 0, so nothing will be rendered. */
111      *xmin = 0.0f;
112      *xmax = 0.0f;
113      *ymin = 0.0f;
114      *ymax = 0.0f;
115   }
116}
117
118#endif /* INTEL_GUARDBAND_H */
119