intel/nir: Allow splitting a single load into up to 32 loads
[mesa.git] / src / intel / isl / isl_gen9.c
1 /*
2 * Copyright 2015 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
24 #include "isl_gen8.h"
25 #include "isl_gen9.h"
26 #include "isl_priv.h"
27
28 /**
29 * Calculate the surface's subimage alignment, in units of surface samples,
30 * for the standard tiling formats Yf and Ys.
31 */
32 static void
33 gen9_calc_std_image_alignment_sa(const struct isl_device *dev,
34 const struct isl_surf_init_info *restrict info,
35 enum isl_tiling tiling,
36 enum isl_msaa_layout msaa_layout,
37 struct isl_extent3d *align_sa)
38 {
39 const struct isl_format_layout *fmtl = isl_format_get_layout(info->format);
40
41 assert(isl_tiling_is_std_y(tiling));
42
43 const uint32_t bpb = fmtl->bpb;
44 const uint32_t is_Ys = tiling == ISL_TILING_Ys;
45
46 switch (info->dim) {
47 case ISL_SURF_DIM_1D:
48 /* See the Skylake BSpec > Memory Views > Common Surface Formats > Surface
49 * Layout and Tiling > 1D Surfaces > 1D Alignment Requirements.
50 */
51 *align_sa = (struct isl_extent3d) {
52 .w = 1 << (12 - (ffs(bpb) - 4) + (4 * is_Ys)),
53 .h = 1,
54 .d = 1,
55 };
56 return;
57 case ISL_SURF_DIM_2D:
58 /* See the Skylake BSpec > Memory Views > Common Surface Formats >
59 * Surface Layout and Tiling > 2D Surfaces > 2D/CUBE Alignment
60 * Requirements.
61 */
62 *align_sa = (struct isl_extent3d) {
63 .w = 1 << (6 - ((ffs(bpb) - 4) / 2) + (4 * is_Ys)),
64 .h = 1 << (6 - ((ffs(bpb) - 3) / 2) + (4 * is_Ys)),
65 .d = 1,
66 };
67
68 if (is_Ys) {
69 /* FINISHME(chadv): I don't trust this code. Untested. */
70 isl_finishme("%s:%s: [SKL+] multisample TileYs", __FILE__, __func__);
71
72 switch (msaa_layout) {
73 case ISL_MSAA_LAYOUT_NONE:
74 case ISL_MSAA_LAYOUT_INTERLEAVED:
75 break;
76 case ISL_MSAA_LAYOUT_ARRAY:
77 align_sa->w >>= (ffs(info->samples) - 0) / 2;
78 align_sa->h >>= (ffs(info->samples) - 1) / 2;
79 break;
80 }
81 }
82 return;
83
84 case ISL_SURF_DIM_3D:
85 /* See the Skylake BSpec > Memory Views > Common Surface Formats > Surface
86 * Layout and Tiling > 1D Surfaces > 1D Alignment Requirements.
87 */
88 *align_sa = (struct isl_extent3d) {
89 .w = 1 << (4 - ((ffs(bpb) - 2) / 3) + (4 * is_Ys)),
90 .h = 1 << (4 - ((ffs(bpb) - 4) / 3) + (2 * is_Ys)),
91 .d = 1 << (4 - ((ffs(bpb) - 3) / 3) + (2 * is_Ys)),
92 };
93 return;
94 }
95
96 unreachable("bad isl_surface_type");
97 }
98
99 void
100 isl_gen9_choose_image_alignment_el(const struct isl_device *dev,
101 const struct isl_surf_init_info *restrict info,
102 enum isl_tiling tiling,
103 enum isl_dim_layout dim_layout,
104 enum isl_msaa_layout msaa_layout,
105 struct isl_extent3d *image_align_el)
106 {
107 /* Handled by isl_choose_image_alignment_el */
108 assert(info->format != ISL_FORMAT_HIZ);
109
110 const struct isl_format_layout *fmtl = isl_format_get_layout(info->format);
111 if (fmtl->txc == ISL_TXC_CCS) {
112 /* Sky Lake PRM Vol. 7, "MCS Buffer for Render Target(s)" (p. 632):
113 *
114 * "Mip-mapped and arrayed surfaces are supported with MCS buffer
115 * layout with these alignments in the RT space: Horizontal
116 * Alignment = 128 and Vertical Alignment = 64."
117 */
118 *image_align_el = isl_extent3d(128 / fmtl->bw, 64 / fmtl->bh, 1);
119 return;
120 }
121
122 /* This BSpec text provides some insight into the hardware's alignment
123 * requirements [Skylake BSpec > Memory Views > Common Surface Formats >
124 * Surface Layout and Tiling > 2D Surfaces]:
125 *
126 * An LOD must be aligned to a cache-line except for some special cases
127 * related to Planar YUV surfaces. In general, the cache-alignment
128 * restriction implies there is a minimum height for an LOD of 4 texels.
129 * So, LODs which are smaller than 4 high are padded.
130 *
131 * From the Skylake BSpec, RENDER_SURFACE_STATE Surface Vertical Alignment:
132 *
133 * - For Sampling Engine and Render Target Surfaces: This field
134 * specifies the vertical alignment requirement in elements for the
135 * surface. [...] An element is defined as a pixel in uncompresed
136 * surface formats, and as a compression block in compressed surface
137 * formats. For MSFMT_DEPTH_STENCIL type multisampled surfaces, an
138 * element is a sample.
139 *
140 * - This field is used for 2D, CUBE, and 3D surface alignment when Tiled
141 * Resource Mode is TRMODE_NONE (Tiled Resource Mode is disabled).
142 * This field is ignored for 1D surfaces and also when Tiled Resource
143 * Mode is not TRMODE_NONE (e.g. Tiled Resource Mode is enabled).
144 *
145 * See the appropriate Alignment table in the "Surface Layout and
146 * Tiling" section under Common Surface Formats for the table of
147 * alignment values for Tiled Resrouces.
148 *
149 * - For uncompressed surfaces, the units of "j" are rows of pixels on
150 * the physical surface. For compressed texture formats, the units of
151 * "j" are in compression blocks, thus each increment in "j" is equal
152 * to h pixels, where h is the height of the compression block in
153 * pixels.
154 *
155 * - Valid Values: VALIGN_4, VALIGN_8, VALIGN_16
156 *
157 * From the Skylake BSpec, RENDER_SURFACE_STATE Surface Horizontal
158 * Alignment:
159 *
160 * - For uncompressed surfaces, the units of "i" are pixels on the
161 * physical surface. For compressed texture formats, the units of "i"
162 * are in compression blocks, thus each increment in "i" is equal to
163 * w pixels, where w is the width of the compression block in pixels.
164 *
165 * - Valid Values: HALIGN_4, HALIGN_8, HALIGN_16
166 */
167
168 if (isl_tiling_is_std_y(tiling)) {
169 struct isl_extent3d image_align_sa;
170 gen9_calc_std_image_alignment_sa(dev, info, tiling, msaa_layout,
171 &image_align_sa);
172
173 *image_align_el = isl_extent3d_sa_to_el(info->format, image_align_sa);
174 return;
175 }
176
177 if (dim_layout == ISL_DIM_LAYOUT_GEN9_1D) {
178 /* See the Skylake BSpec > Memory Views > Common Surface Formats > Surface
179 * Layout and Tiling > 1D Surfaces > 1D Alignment Requirements.
180 */
181 *image_align_el = isl_extent3d(64, 1, 1);
182 return;
183 }
184
185 if (isl_format_is_compressed(info->format)) {
186 /* On Gen9, the meaning of RENDER_SURFACE_STATE's
187 * SurfaceHorizontalAlignment and SurfaceVerticalAlignment changed for
188 * compressed formats. They now indicate a multiple of the compression
189 * block. For example, if the compression mode is ETC2 then HALIGN_4
190 * indicates a horizontal alignment of 16 pixels.
191 *
192 * To avoid wasting memory, choose the smallest alignment possible:
193 * HALIGN_4 and VALIGN_4.
194 */
195 *image_align_el = isl_extent3d(4, 4, 1);
196 return;
197 }
198
199 isl_gen8_choose_image_alignment_el(dev, info, tiling, dim_layout,
200 msaa_layout, image_align_el);
201 }