src/intel/vulkan/genX_pipeline.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 "anv_private.h"
  25
  26 #include "genxml/gen_macros.h"
  27 #include "genxml/genX_pack.h"
  28
  29 static VkResult
  30 compute_pipeline_create(
  31     VkDevice                                    _device,
  32     struct anv_pipeline_cache *                 cache,
  33     const VkComputePipelineCreateInfo*          pCreateInfo,
  34     const VkAllocationCallbacks*                pAllocator,
  35     VkPipeline*                                 pPipeline)
  36 {
  37    ANV_FROM_HANDLE(anv_device, device, _device);
  38    const struct anv_physical_device *physical_device =
  39       &device->instance->physicalDevice;
  40    const struct gen_device_info *devinfo = &physical_device->info;
  41    struct anv_pipeline *pipeline;
  42    VkResult result;
  43
  44    assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO);
  45
  46    pipeline = vk_alloc2(&device->alloc, pAllocator, sizeof(*pipeline), 8,
  47                          VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
  48    if (pipeline == NULL)
  49       return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
  50
  51    pipeline->device = device;
  52    pipeline->layout = anv_pipeline_layout_from_handle(pCreateInfo->layout);
  53
  54    pipeline->blend_state.map = NULL;
  55
  56    result = anv_reloc_list_init(&pipeline->batch_relocs,
  57                                 pAllocator ? pAllocator : &device->alloc);
  58    if (result != VK_SUCCESS) {
  59       vk_free2(&device->alloc, pAllocator, pipeline);
  60       return result;
  61    }
  62    pipeline->batch.next = pipeline->batch.start = pipeline->batch_data;
  63    pipeline->batch.end = pipeline->batch.start + sizeof(pipeline->batch_data);
  64    pipeline->batch.relocs = &pipeline->batch_relocs;
  65
  66    /* When we free the pipeline, we detect stages based on the NULL status
  67     * of various prog_data pointers.  Make them NULL by default.
  68     */
  69    memset(pipeline->shaders, 0, sizeof(pipeline->shaders));
  70
  71    pipeline->vs_simd8 = NO_KERNEL;
  72    pipeline->vs_vec4 = NO_KERNEL;
  73    pipeline->gs_kernel = NO_KERNEL;
  74
  75    pipeline->active_stages = 0;
  76
  77    pipeline->needs_data_cache = false;
  78
  79    assert(pCreateInfo->stage.stage == VK_SHADER_STAGE_COMPUTE_BIT);
  80    ANV_FROM_HANDLE(anv_shader_module, module,  pCreateInfo->stage.module);
  81    result = anv_pipeline_compile_cs(pipeline, cache, pCreateInfo, module,
  82                                     pCreateInfo->stage.pName,
  83                                     pCreateInfo->stage.pSpecializationInfo);
  84    if (result != VK_SUCCESS) {
  85       vk_free2(&device->alloc, pAllocator, pipeline);
  86       return result;
  87    }
  88
  89    const struct brw_cs_prog_data *cs_prog_data = get_cs_prog_data(pipeline);
  90
  91    anv_pipeline_setup_l3_config(pipeline, cs_prog_data->base.total_shared > 0);
  92
  93    uint32_t group_size = cs_prog_data->local_size[0] *
  94       cs_prog_data->local_size[1] * cs_prog_data->local_size[2];
  95    uint32_t remainder = group_size & (cs_prog_data->simd_size - 1);
  96
  97    if (remainder > 0)
  98       pipeline->cs_right_mask = ~0u >> (32 - remainder);
  99    else
 100       pipeline->cs_right_mask = ~0u >> (32 - cs_prog_data->simd_size);
 101
 102    const uint32_t vfe_curbe_allocation =
 103       ALIGN(cs_prog_data->push.per_thread.regs * cs_prog_data->threads +
 104             cs_prog_data->push.cross_thread.regs, 2);
 105
 106    const uint32_t subslices = MAX2(physical_device->subslice_total, 1);
 107
 108    anv_batch_emit(&pipeline->batch, GENX(MEDIA_VFE_STATE), vfe) {
 109       vfe.ScratchSpaceBasePointer = (struct anv_address) {
 110          .bo = anv_scratch_pool_alloc(device, &device->scratch_pool,
 111                                       MESA_SHADER_COMPUTE,
 112                                       cs_prog_data->base.total_scratch),
 113          .offset = 0,
 114       };
 115       vfe.PerThreadScratchSpace  = ffs(cs_prog_data->base.total_scratch / 2048);
 116 #if GEN_GEN > 7
 117       vfe.StackSize              = 0;
 118 #else
 119       vfe.GPGPUMode              = true;
 120 #endif
 121       vfe.MaximumNumberofThreads =
 122          devinfo->max_cs_threads * subslices - 1;
 123       vfe.NumberofURBEntries     = GEN_GEN <= 7 ? 0 : 2;
 124       vfe.ResetGatewayTimer      = true;
 125 #if GEN_GEN <= 8
 126       vfe.BypassGatewayControl   = true;
 127 #endif
 128       vfe.URBEntryAllocationSize = GEN_GEN <= 7 ? 0 : 2;
 129       vfe.CURBEAllocationSize    = vfe_curbe_allocation;
 130    }
 131
 132    *pPipeline = anv_pipeline_to_handle(pipeline);
 133
 134    return VK_SUCCESS;
 135 }
 136
 137 VkResult genX(CreateGraphicsPipelines)(
 138     VkDevice                                    _device,
 139     VkPipelineCache                             pipelineCache,
 140     uint32_t                                    count,
 141     const VkGraphicsPipelineCreateInfo*         pCreateInfos,
 142     const VkAllocationCallbacks*                pAllocator,
 143     VkPipeline*                                 pPipelines)
 144 {
 145    ANV_FROM_HANDLE(anv_pipeline_cache, pipeline_cache, pipelineCache);
 146
 147    VkResult result = VK_SUCCESS;
 148
 149    unsigned i = 0;
 150    for (; i < count; i++) {
 151       result = genX(graphics_pipeline_create)(_device,
 152                                               pipeline_cache,
 153                                               &pCreateInfos[i],
 154                                               pAllocator, &pPipelines[i]);
 155       if (result != VK_SUCCESS) {
 156          for (unsigned j = 0; j < i; j++) {
 157             anv_DestroyPipeline(_device, pPipelines[j], pAllocator);
 158          }
 159
 160          return result;
 161       }
 162    }
 163
 164    return VK_SUCCESS;
 165 }
 166
 167 VkResult genX(CreateComputePipelines)(
 168     VkDevice                                    _device,
 169     VkPipelineCache                             pipelineCache,
 170     uint32_t                                    count,
 171     const VkComputePipelineCreateInfo*          pCreateInfos,
 172     const VkAllocationCallbacks*                pAllocator,
 173     VkPipeline*                                 pPipelines)
 174 {
 175    ANV_FROM_HANDLE(anv_pipeline_cache, pipeline_cache, pipelineCache);
 176
 177    VkResult result = VK_SUCCESS;
 178
 179    unsigned i = 0;
 180    for (; i < count; i++) {
 181       result = compute_pipeline_create(_device, pipeline_cache,
 182                                        &pCreateInfos[i],
 183                                        pAllocator, &pPipelines[i]);
 184       if (result != VK_SUCCESS) {
 185          for (unsigned j = 0; j < i; j++) {
 186             anv_DestroyPipeline(_device, pPipelines[j], pAllocator);
 187          }
 188
 189          return result;
 190       }
 191    }
 192
 193    return VK_SUCCESS;
 194 }