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 VkResult
  30 genX(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    struct anv_pipeline *pipeline;
  39    VkResult result;
  40
  41    assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO);
  42
  43    pipeline = anv_alloc2(&device->alloc, pAllocator, sizeof(*pipeline), 8,
  44                          VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
  45    if (pipeline == NULL)
  46       return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
  47
  48    pipeline->device = device;
  49    pipeline->layout = anv_pipeline_layout_from_handle(pCreateInfo->layout);
  50
  51    pipeline->blend_state.map = NULL;
  52
  53    result = anv_reloc_list_init(&pipeline->batch_relocs,
  54                                 pAllocator ? pAllocator : &device->alloc);
  55    if (result != VK_SUCCESS) {
  56       anv_free2(&device->alloc, pAllocator, pipeline);
  57       return result;
  58    }
  59    pipeline->batch.next = pipeline->batch.start = pipeline->batch_data;
  60    pipeline->batch.end = pipeline->batch.start + sizeof(pipeline->batch_data);
  61    pipeline->batch.relocs = &pipeline->batch_relocs;
  62
  63    /* When we free the pipeline, we detect stages based on the NULL status
  64     * of various prog_data pointers.  Make them NULL by default.
  65     */
  66    memset(pipeline->prog_data, 0, sizeof(pipeline->prog_data));
  67    memset(pipeline->scratch_start, 0, sizeof(pipeline->scratch_start));
  68    memset(pipeline->bindings, 0, sizeof(pipeline->bindings));
  69
  70    pipeline->vs_simd8 = NO_KERNEL;
  71    pipeline->vs_vec4 = NO_KERNEL;
  72    pipeline->gs_kernel = NO_KERNEL;
  73
  74    pipeline->active_stages = 0;
  75    pipeline->total_scratch = 0;
  76
  77    assert(pCreateInfo->stage.stage == VK_SHADER_STAGE_COMPUTE_BIT);
  78    ANV_FROM_HANDLE(anv_shader_module, module,  pCreateInfo->stage.module);
  79    anv_pipeline_compile_cs(pipeline, cache, pCreateInfo, module,
  80                            pCreateInfo->stage.pName,
  81                            pCreateInfo->stage.pSpecializationInfo);
  82
  83    pipeline->use_repclear = false;
  84
  85    const struct brw_cs_prog_data *cs_prog_data = get_cs_prog_data(pipeline);
  86    const struct brw_stage_prog_data *prog_data = &cs_prog_data->base;
  87
  88    unsigned local_id_dwords = cs_prog_data->local_invocation_id_regs * 8;
  89    unsigned push_constant_data_size =
  90       (prog_data->nr_params + local_id_dwords) * 4;
  91    unsigned reg_aligned_constant_size = ALIGN(push_constant_data_size, 32);
  92    unsigned push_constant_regs = reg_aligned_constant_size / 32;
  93
  94    uint32_t group_size = cs_prog_data->local_size[0] *
  95       cs_prog_data->local_size[1] * cs_prog_data->local_size[2];
  96    pipeline->cs_thread_width_max =
  97       DIV_ROUND_UP(group_size, cs_prog_data->simd_size);
  98    uint32_t remainder = group_size & (cs_prog_data->simd_size - 1);
  99
 100    if (remainder > 0)
 101       pipeline->cs_right_mask = ~0u >> (32 - remainder);
 102    else
 103       pipeline->cs_right_mask = ~0u >> (32 - cs_prog_data->simd_size);
 104
 105    const uint32_t vfe_curbe_allocation =
 106       push_constant_regs * pipeline->cs_thread_width_max;
 107
 108    anv_batch_emit(&pipeline->batch, GENX(MEDIA_VFE_STATE),
 109                   .ScratchSpaceBasePointer = pipeline->scratch_start[MESA_SHADER_COMPUTE],
 110                   .PerThreadScratchSpace = ffs(cs_prog_data->base.total_scratch / 2048),
 111 #if GEN_GEN > 7
 112                   .ScratchSpaceBasePointerHigh = 0,
 113                   .StackSize = 0,
 114 #else
 115                   .GPGPUMode = true,
 116 #endif
 117                   .MaximumNumberofThreads = device->info.max_cs_threads - 1,
 118                   .NumberofURBEntries = GEN_GEN <= 7 ? 0 : 2,
 119                   .ResetGatewayTimer = true,
 120 #if GEN_GEN <= 8
 121                   .BypassGatewayControl = true,
 122 #endif
 123                   .URBEntryAllocationSize = GEN_GEN <= 7 ? 0 : 2,
 124                   .CURBEAllocationSize = vfe_curbe_allocation);
 125
 126    *pPipeline = anv_pipeline_to_handle(pipeline);
 127
 128    return VK_SUCCESS;
 129 }