nvk: impl nvk_CmdCopyBuffer

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

src/nouveau/vulkan/meson.build

@@ -55,6 +55,7 @@ nvk_entrypoints = custom_target(
 nvk_deps = [
   dep_libdrm,
   idep_nouveau_ws,
+  idep_nvidia_headers,
   idep_vulkan_runtime,
   idep_vulkan_util,
   idep_vulkan_wsi,

src/nouveau/vulkan/nvk_cmd_buffer.c

@@ -1,10 +1,16 @@
 #include "nvk_cmd_buffer.h"
 
+#include "nvk_buffer.h"
 #include "nvk_device.h"
+#include "nvk_device_memory.h"
 #include "nvk_physical_device.h"
 
+#include "nouveau_bo.h"
 #include "nouveau_push.h"
 
+#include "classes/cla0b5.h"
+#include "push906f.h"
+
 static void
 nvk_destroy_cmd_buffer(struct nvk_cmd_buffer *cmd_buffer)
 {
@@ -210,6 +216,53 @@ nvk_EndCommandBuffer(VkCommandBuffer commandBuffer)
    return VK_SUCCESS;
 }
 
+VKAPI_ATTR void VKAPI_CALL
+nvk_CmdCopyBuffer(VkCommandBuffer commandBuffer,
+                  VkBuffer srcBuffer, VkBuffer dstBuffer,
+                  uint32_t regionCount, const VkBufferCopy* pRegions)
+{
+   VK_FROM_HANDLE(nvk_cmd_buffer, cmd, commandBuffer);
+   VK_FROM_HANDLE(nvk_buffer, src, srcBuffer);
+   VK_FROM_HANDLE(nvk_buffer, dst, dstBuffer);
+   struct nouveau_ws_push *push = cmd->push;
+
+   nouveau_ws_push_ref(push, src->mem->bo, NOUVEAU_WS_BO_RD);
+   nouveau_ws_push_ref(push, dst->mem->bo, NOUVEAU_WS_BO_WR);
+
+   for (unsigned r = 0; r < regionCount; r++) {
+      const VkBufferCopy *region = &pRegions[r];
+      VkDeviceSize dstoff = dst->mem->bo->offset + dst->offset + region->dstOffset;
+      VkDeviceSize srcoff = src->mem->bo->offset + src->offset + region->srcOffset;
+      VkDeviceSize size = region->size;
+
+      while (size) {
+         unsigned bytes = MIN2(size, 1 << 17);
+
+         PUSH_MTHD(push, NVA0B5, OFFSET_IN_UPPER,
+                   NVVAL(NVA0B5, OFFSET_IN_UPPER, UPPER, srcoff >> 32),
+                                 OFFSET_IN_LOWER, srcoff & 0xffffffff,
+
+                                 OFFSET_OUT_UPPER,
+                   NVVAL(NVA0B5, OFFSET_OUT_UPPER, UPPER, dstoff >> 32),
+                                 OFFSET_OUT_LOWER, dstoff & 0xffffffff);
+
+         PUSH_MTHD(push, NVA0B5, LINE_LENGTH_IN, bytes,
+                                 LINE_COUNT, 1);
+
+         PUSH_IMMD(push, NVA0B5, LAUNCH_DMA,
+                   NVDEF(NVA0B5, LAUNCH_DMA, DATA_TRANSFER_TYPE, NON_PIPELINED) |
+                   NVDEF(NVA0B5, LAUNCH_DMA, FLUSH_ENABLE, TRUE) |
+                   NVDEF(NVA0B5, LAUNCH_DMA, SRC_MEMORY_LAYOUT, PITCH) |
+                   NVDEF(NVA0B5, LAUNCH_DMA, DST_MEMORY_LAYOUT, PITCH));
+
+         srcoff += bytes;
+         dstoff += bytes;
+         size -= bytes;
+      }
+   }
+}
+
+
 VKAPI_ATTR void VKAPI_CALL
 nvk_CmdPipelineBarrier2(VkCommandBuffer commandBuffer, const VkDependencyInfo *pDependencyInfo) {
 }