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 #include "genX_pipeline_util.h"
  30
  31 static VkResult
  32 genX(graphics_pipeline_create)(
  33     VkDevice                                    _device,
  34     struct anv_pipeline_cache *                 cache,
  35     const VkGraphicsPipelineCreateInfo*         pCreateInfo,
  36     const VkAllocationCallbacks*                pAllocator,
  37     VkPipeline*                                 pPipeline)
  38 {
  39    ANV_FROM_HANDLE(anv_device, device, _device);
  40    ANV_FROM_HANDLE(anv_render_pass, pass, pCreateInfo->renderPass);
  41    struct anv_subpass *subpass = &pass->subpasses[pCreateInfo->subpass];
  42    struct anv_pipeline *pipeline;
  43    VkResult result;
  44
  45    assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO);
  46
  47    pipeline = vk_alloc2(&device->alloc, pAllocator, sizeof(*pipeline), 8,
  48                          VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
  49    if (pipeline == NULL)
  50       return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
  51
  52    result = anv_pipeline_init(pipeline, device, cache,
  53                               pCreateInfo, pAllocator);
  54    if (result != VK_SUCCESS) {
  55       vk_free2(&device->alloc, pAllocator, pipeline);
  56       return result;
  57    }
  58
  59    assert(pCreateInfo->pVertexInputState);
  60    emit_vertex_input(pipeline, pCreateInfo->pVertexInputState);
  61    assert(pCreateInfo->pRasterizationState);
  62    emit_rs_state(pipeline, pCreateInfo->pRasterizationState,
  63                  pCreateInfo->pMultisampleState, pass, subpass);
  64    emit_ms_state(pipeline, pCreateInfo->pMultisampleState);
  65    emit_ds_state(pipeline, pCreateInfo->pDepthStencilState, pass, subpass);
  66    emit_cb_state(pipeline, pCreateInfo->pColorBlendState,
  67                            pCreateInfo->pMultisampleState);
  68
  69    emit_urb_setup(pipeline);
  70
  71    emit_3dstate_clip(pipeline, pCreateInfo->pViewportState,
  72                      pCreateInfo->pRasterizationState);
  73    emit_3dstate_streamout(pipeline, pCreateInfo->pRasterizationState);
  74
  75 #if 0
  76    /* From gen7_vs_state.c */
  77
  78    /**
  79     * From Graphics BSpec: 3D-Media-GPGPU Engine > 3D Pipeline Stages >
  80     * Geometry > Geometry Shader > State:
  81     *
  82     *     "Note: Because of corruption in IVB:GT2, software needs to flush the
  83     *     whole fixed function pipeline when the GS enable changes value in
  84     *     the 3DSTATE_GS."
  85     *
  86     * The hardware architects have clarified that in this context "flush the
  87     * whole fixed function pipeline" means to emit a PIPE_CONTROL with the "CS
  88     * Stall" bit set.
  89     */
  90    if (!brw->is_haswell && !brw->is_baytrail)
  91       gen7_emit_vs_workaround_flush(brw);
  92 #endif
  93
  94    emit_3dstate_vs(pipeline);
  95    emit_3dstate_gs(pipeline);
  96    emit_3dstate_sbe(pipeline);
  97    emit_3dstate_wm(pipeline, pCreateInfo->pMultisampleState);
  98    emit_3dstate_ps(pipeline);
  99 #if GEN_GEN >= 8
 100    emit_3dstate_ps_extra(pipeline);
 101    emit_3dstate_vf_topology(pipeline);
 102 #endif
 103
 104    *pPipeline = anv_pipeline_to_handle(pipeline);
 105
 106    return VK_SUCCESS;
 107 }
 108
 109 static VkResult
 110 compute_pipeline_create(
 111     VkDevice                                    _device,
 112     struct anv_pipeline_cache *                 cache,
 113     const VkComputePipelineCreateInfo*          pCreateInfo,
 114     const VkAllocationCallbacks*                pAllocator,
 115     VkPipeline*                                 pPipeline)
 116 {
 117    ANV_FROM_HANDLE(anv_device, device, _device);
 118    const struct anv_physical_device *physical_device =
 119       &device->instance->physicalDevice;
 120    const struct gen_device_info *devinfo = &physical_device->info;
 121    struct anv_pipeline *pipeline;
 122    VkResult result;
 123
 124    assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO);
 125
 126    pipeline = vk_alloc2(&device->alloc, pAllocator, sizeof(*pipeline), 8,
 127                          VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
 128    if (pipeline == NULL)
 129       return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
 130
 131    pipeline->device = device;
 132    pipeline->layout = anv_pipeline_layout_from_handle(pCreateInfo->layout);
 133
 134    pipeline->blend_state.map = NULL;
 135
 136    result = anv_reloc_list_init(&pipeline->batch_relocs,
 137                                 pAllocator ? pAllocator : &device->alloc);
 138    if (result != VK_SUCCESS) {
 139       vk_free2(&device->alloc, pAllocator, pipeline);
 140       return result;
 141    }
 142    pipeline->batch.next = pipeline->batch.start = pipeline->batch_data;
 143    pipeline->batch.end = pipeline->batch.start + sizeof(pipeline->batch_data);
 144    pipeline->batch.relocs = &pipeline->batch_relocs;
 145
 146    /* When we free the pipeline, we detect stages based on the NULL status
 147     * of various prog_data pointers.  Make them NULL by default.
 148     */
 149    memset(pipeline->shaders, 0, sizeof(pipeline->shaders));
 150
 151    pipeline->active_stages = 0;
 152
 153    pipeline->needs_data_cache = false;
 154
 155    assert(pCreateInfo->stage.stage == VK_SHADER_STAGE_COMPUTE_BIT);
 156    ANV_FROM_HANDLE(anv_shader_module, module,  pCreateInfo->stage.module);
 157    result = anv_pipeline_compile_cs(pipeline, cache, pCreateInfo, module,
 158                                     pCreateInfo->stage.pName,
 159                                     pCreateInfo->stage.pSpecializationInfo);
 160    if (result != VK_SUCCESS) {
 161       vk_free2(&device->alloc, pAllocator, pipeline);
 162       return result;
 163    }
 164
 165    const struct brw_cs_prog_data *cs_prog_data = get_cs_prog_data(pipeline);
 166
 167    anv_pipeline_setup_l3_config(pipeline, cs_prog_data->base.total_shared > 0);
 168
 169    uint32_t group_size = cs_prog_data->local_size[0] *
 170       cs_prog_data->local_size[1] * cs_prog_data->local_size[2];
 171    uint32_t remainder = group_size & (cs_prog_data->simd_size - 1);
 172
 173    if (remainder > 0)
 174       pipeline->cs_right_mask = ~0u >> (32 - remainder);
 175    else
 176       pipeline->cs_right_mask = ~0u >> (32 - cs_prog_data->simd_size);
 177
 178    const uint32_t vfe_curbe_allocation =
 179       ALIGN(cs_prog_data->push.per_thread.regs * cs_prog_data->threads +
 180             cs_prog_data->push.cross_thread.regs, 2);
 181
 182    const uint32_t subslices = MAX2(physical_device->subslice_total, 1);
 183
 184    anv_batch_emit(&pipeline->batch, GENX(MEDIA_VFE_STATE), vfe) {
 185       vfe.ScratchSpaceBasePointer = (struct anv_address) {
 186          .bo = anv_scratch_pool_alloc(device, &device->scratch_pool,
 187                                       MESA_SHADER_COMPUTE,
 188                                       cs_prog_data->base.total_scratch),
 189          .offset = 0,
 190       };
 191       vfe.PerThreadScratchSpace  = ffs(cs_prog_data->base.total_scratch / 2048);
 192 #if GEN_GEN > 7
 193       vfe.StackSize              = 0;
 194 #else
 195       vfe.GPGPUMode              = true;
 196 #endif
 197       vfe.MaximumNumberofThreads =
 198          devinfo->max_cs_threads * subslices - 1;
 199       vfe.NumberofURBEntries     = GEN_GEN <= 7 ? 0 : 2;
 200       vfe.ResetGatewayTimer      = true;
 201 #if GEN_GEN <= 8
 202       vfe.BypassGatewayControl   = true;
 203 #endif
 204       vfe.URBEntryAllocationSize = GEN_GEN <= 7 ? 0 : 2;
 205       vfe.CURBEAllocationSize    = vfe_curbe_allocation;
 206    }
 207
 208    *pPipeline = anv_pipeline_to_handle(pipeline);
 209
 210    return VK_SUCCESS;
 211 }
 212
 213 VkResult genX(CreateGraphicsPipelines)(
 214     VkDevice                                    _device,
 215     VkPipelineCache                             pipelineCache,
 216     uint32_t                                    count,
 217     const VkGraphicsPipelineCreateInfo*         pCreateInfos,
 218     const VkAllocationCallbacks*                pAllocator,
 219     VkPipeline*                                 pPipelines)
 220 {
 221    ANV_FROM_HANDLE(anv_pipeline_cache, pipeline_cache, pipelineCache);
 222
 223    VkResult result = VK_SUCCESS;
 224
 225    unsigned i = 0;
 226    for (; i < count; i++) {
 227       result = genX(graphics_pipeline_create)(_device,
 228                                               pipeline_cache,
 229                                               &pCreateInfos[i],
 230                                               pAllocator, &pPipelines[i]);
 231       if (result != VK_SUCCESS) {
 232          for (unsigned j = 0; j < i; j++) {
 233             anv_DestroyPipeline(_device, pPipelines[j], pAllocator);
 234          }
 235
 236          return result;
 237       }
 238    }
 239
 240    return VK_SUCCESS;
 241 }
 242
 243 VkResult genX(CreateComputePipelines)(
 244     VkDevice                                    _device,
 245     VkPipelineCache                             pipelineCache,
 246     uint32_t                                    count,
 247     const VkComputePipelineCreateInfo*          pCreateInfos,
 248     const VkAllocationCallbacks*                pAllocator,
 249     VkPipeline*                                 pPipelines)
 250 {
 251    ANV_FROM_HANDLE(anv_pipeline_cache, pipeline_cache, pipelineCache);
 252
 253    VkResult result = VK_SUCCESS;
 254
 255    unsigned i = 0;
 256    for (; i < count; i++) {
 257       result = compute_pipeline_create(_device, pipeline_cache,
 258                                        &pCreateInfos[i],
 259                                        pAllocator, &pPipelines[i]);
 260       if (result != VK_SUCCESS) {
 261          for (unsigned j = 0; j < i; j++) {
 262             anv_DestroyPipeline(_device, pPipelines[j], pAllocator);
 263          }
 264
 265          return result;
 266       }
 267    }
 268
 269    return VK_SUCCESS;
 270 }