anv: narrow flushing of the render target to buffer writes

In commit 9a7b319903 ("anv/query: flush render target before
copying results") we tracked all the render target writes to apply a
flushes in the vkCopyQueryResults(). But we can narrow this down to
only when we write a buffer (which is the only input of
vkCopyQueryResults).

v2: Drop newer render target write flags introduce by 1952fd8d2c
    ("anv: Implement VK_EXT_conditional_rendering for gen 7.5+")

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net> (v1)

src/intel/vulkan/anv_blorp.c

@@ -475,6 +475,8 @@ void anv_CmdCopyImageToBuffer(
 
    copy_buffer_to_image(cmd_buffer, dst_buffer, src_image, srcImageLayout,
                         regionCount, pRegions, false);
+
+   cmd_buffer->state.pending_pipe_bits |= ANV_PIPE_RENDER_TARGET_BUFFER_WRITES;
 }
 
 static bool
@@ -682,6 +684,8 @@ void anv_CmdCopyBuffer(
    }
 
    blorp_batch_finish(&batch);
+
+   cmd_buffer->state.pending_pipe_bits |= ANV_PIPE_RENDER_TARGET_BUFFER_WRITES;
 }
 
 void anv_CmdUpdateBuffer(
@@ -737,6 +741,8 @@ void anv_CmdUpdateBuffer(
    }
 
    blorp_batch_finish(&batch);
+
+   cmd_buffer->state.pending_pipe_bits |= ANV_PIPE_RENDER_TARGET_BUFFER_WRITES;
 }
 
 void anv_CmdFillBuffer(
@@ -824,6 +830,8 @@ void anv_CmdFillBuffer(
    }
 
    blorp_batch_finish(&batch);
+
+   cmd_buffer->state.pending_pipe_bits |= ANV_PIPE_RENDER_TARGET_BUFFER_WRITES;
 }
 
 void anv_CmdClearColorImage(

src/intel/vulkan/anv_private.h

@@ -1790,11 +1790,12 @@ enum anv_pipe_bits {
    ANV_PIPE_NEEDS_CS_STALL_BIT               = (1 << 21),
 
    /* This bit does not exist directly in PIPE_CONTROL. It means that render
-    * target operations are ongoing. Some operations like copies on the
-    * command streamer might need to be aware of this to trigger the
-    * appropriate stall before they can proceed with the copy.
+    * target operations related to transfer commands with VkBuffer as
+    * destination are ongoing. Some operations like copies on the command
+    * streamer might need to be aware of this to trigger the appropriate stall
+    * before they can proceed with the copy.
     */
-   ANV_PIPE_RENDER_TARGET_WRITES              = (1 << 22),
+   ANV_PIPE_RENDER_TARGET_BUFFER_WRITES      = (1 << 22),
 };
 
 #define ANV_PIPE_FLUSH_BITS ( \

src/intel/vulkan/genX_blorp_exec.c

@@ -268,5 +268,4 @@ genX(blorp_exec)(struct blorp_batch *batch,
    cmd_buffer->state.gfx.vb_dirty = ~0;
    cmd_buffer->state.gfx.dirty = ~0;
    cmd_buffer->state.push_constants_dirty = ~0;
-   cmd_buffer->state.pending_pipe_bits |= ANV_PIPE_RENDER_TARGET_WRITES;
 }

src/intel/vulkan/genX_cmd_buffer.c

@@ -1811,7 +1811,7 @@ genX(cmd_buffer_apply_pipe_flushes)(struct anv_cmd_buffer *cmd_buffer)
        * saying that render target writes are ongoing.
        */
       if (bits & ANV_PIPE_RENDER_TARGET_CACHE_FLUSH_BIT)
-         bits &= ~(ANV_PIPE_RENDER_TARGET_WRITES);
+         bits &= ~(ANV_PIPE_RENDER_TARGET_BUFFER_WRITES);
 
       bits &= ~(ANV_PIPE_FLUSH_BITS | ANV_PIPE_CS_STALL_BIT);
    }
@@ -2828,8 +2828,6 @@ void genX(CmdDraw)(
       prim.StartInstanceLocation    = firstInstance;
       prim.BaseVertexLocation       = 0;
    }
-
-   cmd_buffer->state.pending_pipe_bits |= ANV_PIPE_RENDER_TARGET_WRITES;
 }
 
 void genX(CmdDrawIndexed)(
@@ -2873,8 +2871,6 @@ void genX(CmdDrawIndexed)(
       prim.StartInstanceLocation    = firstInstance;
       prim.BaseVertexLocation       = vertexOffset;
    }
-
-   cmd_buffer->state.pending_pipe_bits |= ANV_PIPE_RENDER_TARGET_WRITES;
 }
 
 /* Auto-Draw / Indirect Registers */
@@ -3012,8 +3008,6 @@ void genX(CmdDrawIndirect)(
 
       offset += stride;
    }
-
-   cmd_buffer->state.pending_pipe_bits |= ANV_PIPE_RENDER_TARGET_WRITES;
 }
 
 void genX(CmdDrawIndexedIndirect)(
@@ -3057,8 +3051,6 @@ void genX(CmdDrawIndexedIndirect)(
 
       offset += stride;
    }
-
-   cmd_buffer->state.pending_pipe_bits |= ANV_PIPE_RENDER_TARGET_WRITES;
 }
 
 #define TMP_DRAW_COUNT_REG MI_ALU_REG14
@@ -3218,8 +3210,6 @@ void genX(CmdDrawIndirectCountKHR)(
 
       offset += stride;
    }
-
-   cmd_buffer->state.pending_pipe_bits |= ANV_PIPE_RENDER_TARGET_WRITES;
 }
 
 void genX(CmdDrawIndexedIndirectCountKHR)(
@@ -3280,8 +3270,6 @@ void genX(CmdDrawIndexedIndirectCountKHR)(
 
       offset += stride;
    }
-
-   cmd_buffer->state.pending_pipe_bits |= ANV_PIPE_RENDER_TARGET_WRITES;
 }
 
 static VkResult

src/intel/vulkan/genX_gpu_memcpy.c

@@ -299,5 +299,4 @@ genX(cmd_buffer_so_memcpy)(struct anv_cmd_buffer *cmd_buffer,
    }
 
    cmd_buffer->state.gfx.dirty |= ANV_CMD_DIRTY_PIPELINE;
-   cmd_buffer->state.pending_pipe_bits |= ANV_PIPE_RENDER_TARGET_WRITES;
 }

src/intel/vulkan/genX_query.c

@@ -733,7 +733,7 @@ void genX(CmdCopyQueryPoolResults)(
     * to ensure proper ordering of the commands from the 3d pipe and the
     * command streamer.
     */
-   if (cmd_buffer->state.pending_pipe_bits & ANV_PIPE_RENDER_TARGET_WRITES) {
+   if (cmd_buffer->state.pending_pipe_bits & ANV_PIPE_RENDER_TARGET_BUFFER_WRITES) {
       cmd_buffer->state.pending_pipe_bits |=
          ANV_PIPE_RENDER_TARGET_CACHE_FLUSH_BIT;
    }