Module: Mesa Branch: staging/23.1 Commit: 0d41f0369f60e077233d256e57dacd9e3e387283 URL: http://cgit.freedesktop.org/mesa/mesa/commit/?id=0d41f0369f60e077233d256e57dacd9e3e387283
Author: Lionel Landwerlin <[email protected]> Date: Thu Jun 15 14:04:14 2023 +0300 anv: track buffer writes from shaders for query results writes In the following sequence : - write buffer B with a shader - barrier on buffer from shader-write to transfer - vkCmdCopyQueryPoolResults to buffer B The barrier should take care of ordering things between the shader writes and vkCmdCopyQueryPoolResults. The problem is that vkCmdCopyQueryPoolResults runs on the command streamer and that is not coherent or synchronized in the same way as shaders. This change marks the barrier has potentially containing pending buffer writes for queries so that we can insert the necessary flush for vkCmdCopyQueryPoolResults later. Signed-off-by: Lionel Landwerlin <[email protected]> Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/9013 Cc: mesa-stable Reviewed-by: Rohan Garg <[email protected]> Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23675> (cherry picked from commit cab849562517631db7303f0bdf08d21e60f59ad8) --- .pick_status.json | 2 +- src/intel/vulkan/genX_cmd_buffer.c | 51 ++++++++++++++++++++++++++++++++++++++ 2 files changed, 52 insertions(+), 1 deletion(-) diff --git a/.pick_status.json b/.pick_status.json index e760e44a5e7..2d9eea5a12c 100644 --- a/.pick_status.json +++ b/.pick_status.json @@ -571,7 +571,7 @@ "description": "anv: track buffer writes from shaders for query results writes", "nominated": true, "nomination_type": 0, - "resolution": 0, + "resolution": 1, "main_sha": null, "because_sha": null }, diff --git a/src/intel/vulkan/genX_cmd_buffer.c b/src/intel/vulkan/genX_cmd_buffer.c index 5adc91a6034..b7fe5627aab 100644 --- a/src/intel/vulkan/genX_cmd_buffer.c +++ b/src/intel/vulkan/genX_cmd_buffer.c @@ -4065,6 +4065,37 @@ genX(CmdExecuteCommands)( } } +static inline bool +stage_is_shader(const VkPipelineStageFlags2 stage) +{ + return (stage & (VK_PIPELINE_STAGE_2_VERTEX_SHADER_BIT | + VK_PIPELINE_STAGE_2_TESSELLATION_CONTROL_SHADER_BIT | + VK_PIPELINE_STAGE_2_TESSELLATION_EVALUATION_SHADER_BIT | + VK_PIPELINE_STAGE_2_GEOMETRY_SHADER_BIT | + VK_PIPELINE_STAGE_2_FRAGMENT_SHADER_BIT | + VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT | + VK_PIPELINE_STAGE_2_ALL_GRAPHICS_BIT | + VK_PIPELINE_STAGE_2_ALL_COMMANDS_BIT | + VK_PIPELINE_STAGE_2_RAY_TRACING_SHADER_BIT_KHR | + VK_PIPELINE_STAGE_2_TASK_SHADER_BIT_EXT | + VK_PIPELINE_STAGE_2_MESH_SHADER_BIT_EXT)); +} + +static inline bool +stage_is_transfer(const VkPipelineStageFlags2 stage) +{ + return (stage & (VK_PIPELINE_STAGE_2_ALL_COMMANDS_BIT | + VK_PIPELINE_STAGE_2_TRANSFER_BIT)); +} + +static inline bool +mask_is_shader_write(const VkAccessFlags2 access) +{ + return (access & (VK_ACCESS_2_SHADER_WRITE_BIT | + VK_ACCESS_2_MEMORY_WRITE_BIT | + VK_ACCESS_2_SHADER_STORAGE_WRITE_BIT)); +} + static void cmd_buffer_barrier(struct anv_cmd_buffer *cmd_buffer, const VkDependencyInfo *dep_info, @@ -4083,11 +4114,31 @@ cmd_buffer_barrier(struct anv_cmd_buffer *cmd_buffer, for (uint32_t i = 0; i < dep_info->memoryBarrierCount; i++) { src_flags |= dep_info->pMemoryBarriers[i].srcAccessMask; dst_flags |= dep_info->pMemoryBarriers[i].dstAccessMask; + + /* Shader writes to buffers that could then be written by a transfer + * command (including queries). + */ + if (stage_is_shader(dep_info->pMemoryBarriers[i].srcStageMask) && + mask_is_shader_write(dep_info->pMemoryBarriers[i].srcAccessMask) && + stage_is_transfer(dep_info->pMemoryBarriers[i].dstStageMask)) { + cmd_buffer->state.pending_query_bits |= + ANV_QUERY_COMPUTE_WRITES_PENDING_BITS; + } } for (uint32_t i = 0; i < dep_info->bufferMemoryBarrierCount; i++) { src_flags |= dep_info->pBufferMemoryBarriers[i].srcAccessMask; dst_flags |= dep_info->pBufferMemoryBarriers[i].dstAccessMask; + + /* Shader writes to buffers that could then be written by a transfer + * command (including queries). + */ + if (stage_is_shader(dep_info->pBufferMemoryBarriers[i].srcStageMask) && + mask_is_shader_write(dep_info->pBufferMemoryBarriers[i].srcAccessMask) && + stage_is_transfer(dep_info->pBufferMemoryBarriers[i].dstStageMask)) { + cmd_buffer->state.pending_query_bits |= + ANV_QUERY_COMPUTE_WRITES_PENDING_BITS; + } } for (uint32_t i = 0; i < dep_info->imageMemoryBarrierCount; i++) {
