src/compiler/nir/nir_gather_xfb_info.c

   1 /*
   2  * Copyright © 2018 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 "nir_xfb_info.h"
  25
  26 #include <util/u_math.h>
  27
  28 static void
  29 add_var_xfb_outputs(nir_xfb_info *xfb,
  30                     nir_variable *var,
  31                     unsigned buffer,
  32                     unsigned *location,
  33                     unsigned *offset,
  34                     const struct glsl_type *type)
  35 {
  36    /* If this type contains a 64-bit value, align to 8 bytes */
  37    if (glsl_type_contains_64bit(type))
  38       *offset = ALIGN_POT(*offset, 8);
  39
  40    if (glsl_type_is_array_or_matrix(type) && !var->data.compact) {
  41       unsigned length = glsl_get_length(type);
  42       const struct glsl_type *child_type = glsl_get_array_element(type);
  43       for (unsigned i = 0; i < length; i++)
  44          add_var_xfb_outputs(xfb, var, buffer, location, offset, child_type);
  45    } else if (glsl_type_is_struct(type)) {
  46       unsigned length = glsl_get_length(type);
  47       for (unsigned i = 0; i < length; i++) {
  48          const struct glsl_type *child_type = glsl_get_struct_field(type, i);
  49          add_var_xfb_outputs(xfb, var, buffer, location, offset, child_type);
  50       }
  51    } else {
  52       assert(buffer < NIR_MAX_XFB_BUFFERS);
  53       if (xfb->buffers_written & (1 << buffer)) {
  54          assert(xfb->strides[buffer] == var->data.xfb_stride);
  55          assert(xfb->buffer_to_stream[buffer] == var->data.stream);
  56       } else {
  57          xfb->buffers_written |= (1 << buffer);
  58          xfb->strides[buffer] = var->data.xfb_stride;
  59          xfb->buffer_to_stream[buffer] = var->data.stream;
  60       }
  61
  62       assert(var->data.stream < NIR_MAX_XFB_STREAMS);
  63       xfb->streams_written |= (1 << var->data.stream);
  64
  65       unsigned comp_slots;
  66       if (var->data.compact) {
  67          /* This only happens for clip/cull which are float arrays */
  68          assert(glsl_without_array(type) == glsl_float_type());
  69          assert(var->data.location == VARYING_SLOT_CLIP_DIST0 ||
  70                 var->data.location == VARYING_SLOT_CLIP_DIST1);
  71          comp_slots = glsl_get_length(type);
  72       } else {
  73          comp_slots = glsl_get_component_slots(type);
  74
  75          unsigned attrib_slots = DIV_ROUND_UP(comp_slots, 4);
  76          assert(attrib_slots == glsl_count_attribute_slots(type, false));
  77
  78          /* Ensure that we don't have, for instance, a dvec2 with a
  79           * location_frac of 2 which would make it crass a location boundary
  80           * even though it fits in a single slot.  However, you can have a
  81           * dvec3 which crosses the slot boundary with a location_frac of 2.
  82           */
  83          assert(DIV_ROUND_UP(var->data.location_frac + comp_slots, 4) ==
  84                 attrib_slots);
  85       }
  86
  87       assert(var->data.location_frac + comp_slots <= 8);
  88       uint8_t comp_mask = ((1 << comp_slots) - 1) << var->data.location_frac;
  89
  90       while (comp_mask) {
  91          nir_xfb_output_info *output = &xfb->outputs[xfb->output_count++];
  92
  93          output->buffer = buffer;
  94          output->offset = *offset;
  95          output->location = *location;
  96          output->component_mask = comp_mask & 0xf;
  97
  98          *offset += util_bitcount(output->component_mask) * 4;
  99          (*location)++;
 100          comp_mask >>= 4;
 101       }
 102    }
 103 }
 104
 105 static int
 106 compare_xfb_output_offsets(const void *_a, const void *_b)
 107 {
 108    const nir_xfb_output_info *a = _a, *b = _b;
 109    return a->offset - b->offset;
 110 }
 111
 112 nir_xfb_info *
 113 nir_gather_xfb_info(const nir_shader *shader, void *mem_ctx)
 114 {
 115    assert(shader->info.stage == MESA_SHADER_VERTEX ||
 116           shader->info.stage == MESA_SHADER_TESS_EVAL ||
 117           shader->info.stage == MESA_SHADER_GEOMETRY);
 118
 119    /* Compute the number of outputs we have.  This is simply the number of
 120     * cumulative locations consumed by all the variables.  If a location is
 121     * represented by multiple variables, then they each count separately in
 122     * number of outputs.  This is only an estimate as some variables may have
 123     * an xfb_buffer but not an output so it may end up larger than we need but
 124     * it should be good enough for allocation.
 125     */
 126    unsigned num_outputs = 0;
 127    nir_foreach_variable(var, &shader->outputs) {
 128       if (var->data.explicit_xfb_buffer)
 129          num_outputs += glsl_count_attribute_slots(var->type, false);
 130    }
 131    if (num_outputs == 0)
 132       return NULL;
 133
 134    nir_xfb_info *xfb = rzalloc_size(mem_ctx, nir_xfb_info_size(num_outputs));
 135
 136    /* Walk the list of outputs and add them to the array */
 137    nir_foreach_variable(var, &shader->outputs) {
 138       if (!var->data.explicit_xfb_buffer)
 139          continue;
 140
 141       unsigned location = var->data.location;
 142
 143       /* In order to know if we have a array of blocks can't be done just by
 144        * checking if we have an interface type and is an array, because due
 145        * splitting we could end on a case were we received a split struct
 146        * that contains an array.
 147        */
 148       bool is_array_block = var->interface_type != NULL &&
 149          glsl_type_is_array(var->type) &&
 150          glsl_without_array(var->type) == glsl_get_bare_type(var->interface_type);
 151
 152       if (var->data.explicit_offset && !is_array_block) {
 153          unsigned offset = var->data.offset;
 154          add_var_xfb_outputs(xfb, var, var->data.xfb_buffer,
 155                              &location, &offset, var->type);
 156       } else if (is_array_block) {
 157          assert(glsl_type_is_struct(var->interface_type));
 158
 159          unsigned aoa_size = glsl_get_aoa_size(var->type);
 160          const struct glsl_type *itype = var->interface_type;
 161          unsigned nfields = glsl_get_length(itype);
 162          for (unsigned b = 0; b < aoa_size; b++) {
 163             for (unsigned f = 0; f < nfields; f++) {
 164                int foffset = glsl_get_struct_field_offset(itype, f);
 165                const struct glsl_type *ftype = glsl_get_struct_field(itype, f);
 166                if (foffset < 0) {
 167                   location += glsl_count_attribute_slots(ftype, false);
 168                   continue;
 169                }
 170
 171                unsigned offset = foffset;
 172                add_var_xfb_outputs(xfb, var, var->data.xfb_buffer + b,
 173                                    &location, &offset, ftype);
 174             }
 175          }
 176       }
 177    }
 178
 179    /* Everything is easier in the state setup code if the list is sorted in
 180     * order of output offset.
 181     */
 182    qsort(xfb->outputs, xfb->output_count, sizeof(xfb->outputs[0]),
 183          compare_xfb_output_offsets);
 184
 185 #ifndef NDEBUG
 186    /* Finally, do a sanity check */
 187    unsigned max_offset[NIR_MAX_XFB_BUFFERS] = {0};
 188    for (unsigned i = 0; i < xfb->output_count; i++) {
 189       assert(xfb->outputs[i].offset >= max_offset[xfb->outputs[i].buffer]);
 190       assert(xfb->outputs[i].component_mask != 0);
 191       unsigned slots = util_bitcount(xfb->outputs[i].component_mask);
 192       max_offset[xfb->outputs[i].buffer] = xfb->outputs[i].offset + slots * 4;
 193    }
 194 #endif
 195
 196    return xfb;
 197 }