src/panfrost/encoder/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         /* First of all, all 6 values are off-by-one (strictly positive).
  54          * Account for that, first by ensuring all values are strictly positive
  55          * and then by offsetting */
  56
  57         assert(num_x > 0);
  58         assert(num_y > 0);
  59         assert(num_z > 0);
  60
  61         assert(size_x > 0);
  62         assert(size_y > 0);
  63         assert(size_z > 0);
  64
  65         num_x = MALI_POSITIVE(num_x);
  66         num_y = MALI_POSITIVE(num_y);
  67         num_z = MALI_POSITIVE(num_z);
  68
  69         size_x = MALI_POSITIVE(size_x);
  70         size_y = MALI_POSITIVE(size_y);
  71         size_z = MALI_POSITIVE(size_z);
  72
  73         /* Next up is to pack in order */
  74
  75         uint32_t packed = 0;
  76
  77         /* The values needing packing, in order, and the corresponding shifts.
  78          * Indicies into shift are off-by-one to make the logic easier */
  79
  80         unsigned shifts[7] = { 0 };
  81         unsigned values[6] = { size_x, size_y, size_z, num_x, num_y, num_z };
  82
  83         for (unsigned i = 0; i < 6; ++i) {
  84                 /* OR it in, shifting as required */
  85                 packed |= (values[i] << shifts[i]);
  86
  87                 /* How many bits did we use? */
  88                 unsigned bit_count = util_logbase2_ceil(values[i] + 1);
  89
  90                 /* Set the next shift accordingly */
  91                 shifts[i + 1] = shifts[i] + bit_count;
  92         }
  93
  94         /* Quirk: for non-instanced graphics, the blob sets workgroups_z_shift
  95          * = 32. This doesn't appear to matter to the hardware, but it's good
  96          * to be bit-identical. */
  97
  98         if (quirk_graphics && (num_z <= 1))
  99                 shifts[5] = 32;
 100
 101         /* Quirk: for graphics, workgroups_x_shift_2 must be at least 2,
 102          * whereas for OpenCL it is simply equal to workgroups_x_shift. For GL
 103          * compute, it seems it might *always* be 2, but this is suspicious and
 104          * needs further investigation. (I'm probably just using GL wrong). */
 105
 106         unsigned shift_2 = shifts[3];
 107
 108         if (quirk_graphics)
 109                 shift_2 = MAX2(shift_2, 2);
 110
 111         /* Pack them in */
 112         uint32_t packed_shifts =
 113                 (shifts[1] << 0) |
 114                 (shifts[2] << 5) |
 115                 (shifts[3] << 10) |
 116                 (shifts[4] << 16) |
 117                 (shifts[5] << 22) |
 118                 (shift_2 << 28);
 119
 120         /* Upload the packed bitfields */
 121         out->invocation_count = packed;
 122         out->invocation_shifts = packed_shifts;
 123
 124         /* TODO: Compute workgroups_x_shift_3 */
 125         out->workgroups_x_shift_3 = shift_2;
 126 }
 127
 128 /* Packs vertex/tiler descriptors simultaneously */
 129 void
 130 panfrost_pack_work_groups_fused(
 131         struct mali_vertex_tiler_prefix *vertex,
 132         struct mali_vertex_tiler_prefix *tiler,
 133         unsigned num_x,
 134         unsigned num_y,
 135         unsigned num_z,
 136         unsigned size_x,
 137         unsigned size_y,
 138         unsigned size_z)
 139 {
 140         panfrost_pack_work_groups_compute(vertex, num_x, num_y, num_z, size_x, size_y, size_z, true);
 141
 142         /* Copy results over */
 143         tiler->invocation_count = vertex->invocation_count;
 144         tiler->invocation_shifts = vertex->invocation_shifts;
 145
 146         /* Set special fields for each */
 147         vertex->workgroups_x_shift_3 = 5;
 148         tiler->workgroups_x_shift_3 = 6;
 149 }
 150