src/panfrost/lib/pan_invocation.c

   1 /*
   2  * Copyright (C) 2019 Collabora, Ltd.
   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 FROM,
  20  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  21  * SOFTWARE.
  22  *
  23  * Authors (Collabora):
  24  *   Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
  25  *
  26  */
  27
  28 #include <assert.h>
  29 #include "util/u_math.h"
  30 #include "pan_encoder.h"
  31
  32 /* Compute shaders are invoked with a gl_NumWorkGroups X/Y/Z triplet. Vertex
  33  * shaders, it turns out, are invoked with the same mechanism, with the triplet
  34  * (1, vertex_count, instance_count).
  35  *
  36  * Alongside this triplet is the gl_WorkGroupSize X/Y/Z triplet.
  37  *
  38  * Unfortunately, the packing for these triplet into the
  39  * mali_vertex_tiler_prefix is a little funky, using a dynamic bitfield. The
  40  * routines here exist to pack this */
  41
  42 void
  43 panfrost_pack_work_groups_compute(
  44         struct mali_vertex_tiler_prefix *out,
  45         unsigned num_x,
  46         unsigned num_y,
  47         unsigned num_z,
  48         unsigned size_x,
  49         unsigned size_y,
  50         unsigned size_z,
  51         bool quirk_graphics)
  52 {
  53         uint32_t packed = 0;
  54
  55         /* The values needing packing, in order, and the corresponding shifts.
  56          * Indicies into shift are off-by-one to make the logic easier */
  57
  58         unsigned shifts[7] = { 0 };
  59
  60         unsigned values[6] = {
  61                 MALI_POSITIVE(size_x),
  62                 MALI_POSITIVE(size_y),
  63                 MALI_POSITIVE(size_z),
  64                 MALI_POSITIVE(num_x),
  65                 MALI_POSITIVE(num_y),
  66                 MALI_POSITIVE(num_z),
  67         };
  68
  69         for (unsigned i = 0; i < 6; ++i) {
  70                 /* OR it in, shifting as required */
  71                 packed |= (values[i] << shifts[i]);
  72
  73                 /* How many bits did we use? */
  74                 unsigned bit_count = util_logbase2_ceil(values[i] + 1);
  75
  76                 /* Set the next shift accordingly */
  77                 shifts[i + 1] = shifts[i] + bit_count;
  78         }
  79
  80         /* Quirk: for non-instanced graphics, the blob sets workgroups_z_shift
  81          * = 32. This doesn't appear to matter to the hardware, but it's good
  82          * to be bit-identical. */
  83
  84         if (quirk_graphics && (num_z <= 1))
  85                 shifts[5] = 32;
  86
  87         /* Quirk: for graphics, workgroups_x_shift_2 must be at least 2,
  88          * whereas for OpenCL it is simply equal to workgroups_x_shift. For GL
  89          * compute, it is always 2 if no barriers are in use, but is equal to
  90          * workgroups_x_shift is barriers are in use. */
  91
  92         unsigned shift_2 = shifts[3];
  93
  94         if (quirk_graphics)
  95                 shift_2 = MAX2(shift_2, 2);
  96
  97         /* Pack them in */
  98         uint32_t packed_shifts =
  99                 (shifts[1] << 0) |
 100                 (shifts[2] << 5) |
 101                 (shifts[3] << 10) |
 102                 (shifts[4] << 16) |
 103                 (shifts[5] << 22) |
 104                 (shift_2 << 28);
 105
 106         /* Upload the packed bitfields */
 107         out->invocation_count = packed;
 108         out->invocation_shifts = packed_shifts;
 109 }
 110
 111 /* Packs vertex/tiler descriptors simultaneously */
 112 void
 113 panfrost_pack_work_groups_fused(
 114         struct mali_vertex_tiler_prefix *vertex,
 115         struct mali_vertex_tiler_prefix *tiler,
 116         unsigned num_x,
 117         unsigned num_y,
 118         unsigned num_z,
 119         unsigned size_x,
 120         unsigned size_y,
 121         unsigned size_z)
 122 {
 123         panfrost_pack_work_groups_compute(vertex, num_x, num_y, num_z, size_x, size_y, size_z, true);
 124
 125         /* Copy results over */
 126         tiler->invocation_count = vertex->invocation_count;
 127         tiler->invocation_shifts = vertex->invocation_shifts;
 128 }
 129