intel/blorp: Add a flag to make blorp not re-emit dept/stencil buffers

In Vulkan, we want to be able to use blorp to perform clears inside of a
render pass.  If blorp stomps the depth/stencil buffers packets then we'll
have to re-emit them.  This gets tricky when secondary command buffers get
involved.  Instead, we'll simply guarantee that the depth and stencil
buffers we pass to blorp (if any) match those already set in the hardware.

Signed-off-by: Jason Ekstrand <[email protected]>
Reviewed-by: Topi Pohjolainen <[email protected]>

4 files changed, 26 insertions, 12 deletions

diff --git a/src/intel/blorp/blorp.c b/src/intel/blorp/blorp.c
index 91513a0413a..08afffe06d1 100644
--- a/src/intel/blorp/blorp.c
+++ b/src/intel/blorp/blorp.c
@@ -45,10 +45,12 @@ blorp_finish(struct blorp_context *blorp)
 
 void
 blorp_batch_init(struct blorp_context *blorp,
-                 struct blorp_batch *batch, void *driver_batch)
+                 struct blorp_batch *batch, void *driver_batch,
+                 enum blorp_batch_flags flags)
 {
    batch->blorp = blorp;
    batch->driver_batch = driver_batch;
+   batch->flags = flags;
 }
 
 void
diff --git a/src/intel/blorp/blorp.h b/src/intel/blorp/blorp.h
index b178ae374d6..0c64d13371b 100644
--- a/src/intel/blorp/blorp.h
+++ b/src/intel/blorp/blorp.h
@@ -67,13 +67,24 @@ void blorp_init(struct blorp_context *blorp, void *driver_ctx,
                 struct isl_device *isl_dev);
 void blorp_finish(struct blorp_context *blorp);
 
+enum blorp_batch_flags {
+   /**
+    * This flag indicates that blorp should *not* re-emit the depth and
+    * stencil buffer packets.  Instead, the driver guarantees that all depth
+    * and stencil images passed in will match what is currently set in the
+    * hardware.
+    */
+   BLORP_BATCH_NO_EMIT_DEPTH_STENCIL = (1 << 0),
+};
+
 struct blorp_batch {
    struct blorp_context *blorp;
    void *driver_batch;
+   enum blorp_batch_flags flags;
 };
 
 void blorp_batch_init(struct blorp_context *blorp, struct blorp_batch *batch,
-                      void *driver_batch);
+                      void *driver_batch, enum blorp_batch_flags flags);
 void blorp_batch_finish(struct blorp_batch *batch);
 
 struct blorp_address {
diff --git a/src/intel/blorp/blorp_genX_exec.h b/src/intel/blorp/blorp_genX_exec.h
index fa5cebf8242..ec0d0223ad3 100644
--- a/src/intel/blorp/blorp_genX_exec.h
+++ b/src/intel/blorp/blorp_genX_exec.h
@@ -1343,7 +1343,8 @@ blorp_exec(struct blorp_batch *batch, const struct blorp_params *params)
 
    blorp_emit_viewport_state(batch, params);
 
-   blorp_emit_depth_stencil_config(batch, params);
+   if (!(batch->flags & BLORP_BATCH_NO_EMIT_DEPTH_STENCIL))
+      blorp_emit_depth_stencil_config(batch, params);
 
    blorp_emit(batch, GENX(3DPRIMITIVE), prim) {
       prim.VertexAccessType = SEQUENTIAL;
diff --git a/src/intel/vulkan/anv_blorp.c b/src/intel/vulkan/anv_blorp.c
index f149f849426..d7a1fd37584 100644
--- a/src/intel/vulkan/anv_blorp.c
+++ b/src/intel/vulkan/anv_blorp.c
@@ -182,7 +182,7 @@ void anv_CmdCopyImage(
    ANV_FROM_HANDLE(anv_image, dst_image, dstImage);
 
    struct blorp_batch batch;
-   blorp_batch_init(&cmd_buffer->device->blorp, &batch, cmd_buffer);
+   blorp_batch_init(&cmd_buffer->device->blorp, &batch, cmd_buffer, 0);
 
    for (unsigned r = 0; r < regionCount; r++) {
       VkOffset3D srcOffset =
@@ -244,7 +244,7 @@ copy_buffer_to_image(struct anv_cmd_buffer *cmd_buffer,
                      bool buffer_to_image)
 {
    struct blorp_batch batch;
-   blorp_batch_init(&cmd_buffer->device->blorp, &batch, cmd_buffer);
+   blorp_batch_init(&cmd_buffer->device->blorp, &batch, cmd_buffer, 0);
 
    struct {
       struct blorp_surf surf;
@@ -404,7 +404,7 @@ void anv_CmdBlitImage(
    }
 
    struct blorp_batch batch;
-   blorp_batch_init(&cmd_buffer->device->blorp, &batch, cmd_buffer);
+   blorp_batch_init(&cmd_buffer->device->blorp, &batch, cmd_buffer, 0);
 
    for (unsigned r = 0; r < regionCount; r++) {
       const VkImageSubresourceLayers *src_res = &pRegions[r].srcSubresource;
@@ -575,7 +575,7 @@ void anv_CmdCopyBuffer(
    ANV_FROM_HANDLE(anv_buffer, dst_buffer, dstBuffer);
 
    struct blorp_batch batch;
-   blorp_batch_init(&cmd_buffer->device->blorp, &batch, cmd_buffer);
+   blorp_batch_init(&cmd_buffer->device->blorp, &batch, cmd_buffer, 0);
 
    for (unsigned r = 0; r < regionCount; r++) {
       uint64_t src_offset = src_buffer->offset + pRegions[r].srcOffset;
@@ -636,7 +636,7 @@ void anv_CmdUpdateBuffer(
    ANV_FROM_HANDLE(anv_buffer, dst_buffer, dstBuffer);
 
    struct blorp_batch batch;
-   blorp_batch_init(&cmd_buffer->device->blorp, &batch, cmd_buffer);
+   blorp_batch_init(&cmd_buffer->device->blorp, &batch, cmd_buffer, 0);
 
    /* We can't quite grab a full block because the state stream needs a
     * little data at the top to build its linked list.
@@ -685,7 +685,7 @@ void anv_CmdFillBuffer(
    struct isl_surf isl_surf;
 
    struct blorp_batch batch;
-   blorp_batch_init(&cmd_buffer->device->blorp, &batch, cmd_buffer);
+   blorp_batch_init(&cmd_buffer->device->blorp, &batch, cmd_buffer, 0);
 
    if (fillSize == VK_WHOLE_SIZE) {
       fillSize = dst_buffer->size - dstOffset;
@@ -767,7 +767,7 @@ void anv_CmdClearColorImage(
    static const bool color_write_disable[4] = { false, false, false, false };
 
    struct blorp_batch batch;
-   blorp_batch_init(&cmd_buffer->device->blorp, &batch, cmd_buffer);
+   blorp_batch_init(&cmd_buffer->device->blorp, &batch, cmd_buffer, 0);
 
    union isl_color_value clear_color;
    memcpy(clear_color.u32, pColor->uint32, sizeof(pColor->uint32));
@@ -857,7 +857,7 @@ void anv_CmdResolveImage(
    ANV_FROM_HANDLE(anv_image, dst_image, dstImage);
 
    struct blorp_batch batch;
-   blorp_batch_init(&cmd_buffer->device->blorp, &batch, cmd_buffer);
+   blorp_batch_init(&cmd_buffer->device->blorp, &batch, cmd_buffer, 0);
 
    for (uint32_t r = 0; r < regionCount; r++) {
       assert(pRegions[r].srcSubresource.aspectMask ==
@@ -902,7 +902,7 @@ anv_cmd_buffer_resolve_subpass(struct anv_cmd_buffer *cmd_buffer)
       return;
 
    struct blorp_batch batch;
-   blorp_batch_init(&cmd_buffer->device->blorp, &batch, cmd_buffer);
+   blorp_batch_init(&cmd_buffer->device->blorp, &batch, cmd_buffer, 0);
 
    for (uint32_t i = 0; i < subpass->color_count; ++i) {
       uint32_t src_att = subpass->color_attachments[i];