radv: Add top of pipe timestamp queries.

Does not fix brokenness with the ready bit.

Signed-off-by: Bas Nieuwenhuizen <basni@google.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>

diff --git a/src/amd/common/sid.h b/src/amd/common/sid.h
index 75ba965..08cdfd7 100644 (file)
--- a/src/amd/common/sid.h
+++ b/src/amd/common/sid.h
@@ -154,6 +154,7 @@
 #define                        COPY_DATA_MEM           1
 #define                 COPY_DATA_PERF          4
 #define                 COPY_DATA_IMM           5
+#define                 COPY_DATA_TIMESTAMP     9
 #define                COPY_DATA_DST_SEL(x)            (((unsigned)(x) & 0xf) << 8)
 #define                COPY_DATA_COUNT_SEL             (1 << 16)
 #define                COPY_DATA_WR_CONFIRM            (1 << 20)

diff --git a/src/amd/vulkan/radv_query.c b/src/amd/vulkan/radv_query.c
index 7e0fd1d..0991c26 100644 (file)
--- a/src/amd/vulkan/radv_query.c
+++ b/src/amd/vulkan/radv_query.c
@@ -1196,21 +1196,35 @@ void radv_CmdWriteTimestamp(
 
        MAYBE_UNUSED unsigned cdw_max = radeon_check_space(cmd_buffer->device->ws, cs, 12);
 
-       if (mec) {
-               radeon_emit(cs, PKT3(PKT3_RELEASE_MEM, 5, 0));
-               radeon_emit(cs, EVENT_TYPE(V_028A90_BOTTOM_OF_PIPE_TS) | EVENT_INDEX(5));
-               radeon_emit(cs, 3 << 29);
-               radeon_emit(cs, query_va);
-               radeon_emit(cs, query_va >> 32);
+       switch(pipelineStage) {
+       case VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT:
+               radeon_emit(cs, PKT3(PKT3_COPY_DATA, 4, 0));
+               radeon_emit(cs, COPY_DATA_COUNT_SEL | COPY_DATA_WR_CONFIRM |
+                               COPY_DATA_SRC_SEL(COPY_DATA_TIMESTAMP) |
+                               COPY_DATA_DST_SEL(V_370_MEM_ASYNC));
                radeon_emit(cs, 0);
                radeon_emit(cs, 0);
-       } else {
-               radeon_emit(cs, PKT3(PKT3_EVENT_WRITE_EOP, 4, 0));
-               radeon_emit(cs, EVENT_TYPE(V_028A90_BOTTOM_OF_PIPE_TS) | EVENT_INDEX(5));
                radeon_emit(cs, query_va);
-               radeon_emit(cs, (3 << 29) | ((query_va >> 32) & 0xFFFF));
-               radeon_emit(cs, 0);
-               radeon_emit(cs, 0);
+               radeon_emit(cs, query_va >> 32);
+               break;
+       default:
+               if (mec) {
+                       radeon_emit(cs, PKT3(PKT3_RELEASE_MEM, 5, 0));
+                       radeon_emit(cs, EVENT_TYPE(V_028A90_BOTTOM_OF_PIPE_TS) | EVENT_INDEX(5));
+                       radeon_emit(cs, 3 << 29);
+                       radeon_emit(cs, query_va);
+                       radeon_emit(cs, query_va >> 32);
+                       radeon_emit(cs, 0);
+                       radeon_emit(cs, 0);
+               } else {
+                       radeon_emit(cs, PKT3(PKT3_EVENT_WRITE_EOP, 4, 0));
+                       radeon_emit(cs, EVENT_TYPE(V_028A90_BOTTOM_OF_PIPE_TS) | EVENT_INDEX(5));
+                       radeon_emit(cs, query_va);
+                       radeon_emit(cs, (3 << 29) | ((query_va >> 32) & 0xFFFF));
+                       radeon_emit(cs, 0);
+                       radeon_emit(cs, 0);
+               }
+               break;
        }
 
        radeon_emit(cs, PKT3(PKT3_WRITE_DATA, 3, 0));