gallium/radeon: add new HUD queries for monitoring the CP

There are even more counters in the CP_STAT register but I think
these ones are enough for now.

v2: only read (and expose) CP_STAT on VI+

Signed-off-by: Samuel Pitoiset <[email protected]>
Reviewed-by: Marek Olšák <[email protected]>

4 files changed, 80 insertions, 3 deletions

diff --git a/src/gallium/drivers/radeon/r600_gpu_load.c b/src/gallium/drivers/radeon/r600_gpu_load.c
index 058006b8261..3b491188fb1 100644
--- a/src/gallium/drivers/radeon/r600_gpu_load.c
+++ b/src/gallium/drivers/radeon/r600_gpu_load.c
@@ -61,6 +61,15 @@
 #define SRBM_STATUS2		0x0e4c
 #define SDMA_BUSY(x)		(((x) >> 5) & 0x1)
 
+#define CP_STAT                 0x8680
+#define PFP_BUSY(x)		(((x) >> 15) & 0x1)
+#define MEQ_BUSY(x)		(((x) >> 16) & 0x1)
+#define ME_BUSY(x)		(((x) >> 17) & 0x1)
+#define SURFACE_SYNC_BUSY(x)	(((x) >> 21) & 0x1)
+#define DMA_BUSY(x)		(((x) >> 22) & 0x1)
+#define SCRATCH_RAM_BUSY(x)	(((x) >> 24) & 0x1)
+#define CE_BUSY(x)		(((x) >> 26) & 0x1)
+
 #define UPDATE_COUNTER(field, mask)					\
 	do {								\
 		if (mask(value))					\
@@ -98,6 +107,19 @@ static void r600_update_mmio_counters(struct r600_common_screen *rscreen,
 
 		UPDATE_COUNTER(sdma, SDMA_BUSY);
 	}
+
+	if (rscreen->chip_class >= VI) {
+		/* CP_STAT */
+		rscreen->ws->read_registers(rscreen->ws, CP_STAT, 1, &value);
+
+		UPDATE_COUNTER(pfp, PFP_BUSY);
+		UPDATE_COUNTER(meq, MEQ_BUSY);
+		UPDATE_COUNTER(me, ME_BUSY);
+		UPDATE_COUNTER(surf_sync, SURFACE_SYNC_BUSY);
+		UPDATE_COUNTER(dma, DMA_BUSY);
+		UPDATE_COUNTER(scratch_ram, SCRATCH_RAM_BUSY);
+		UPDATE_COUNTER(ce, CE_BUSY);
+	}
 }
 
 #undef UPDATE_COUNTER
@@ -223,6 +245,20 @@ static unsigned busy_index_from_type(struct r600_common_screen *rscreen,
 		return BUSY_INDEX(rscreen, cb);
 	case R600_QUERY_GPU_SDMA_BUSY:
 		return BUSY_INDEX(rscreen, sdma);
+	case R600_QUERY_GPU_PFP_BUSY:
+		return BUSY_INDEX(rscreen, pfp);
+	case R600_QUERY_GPU_MEQ_BUSY:
+		return BUSY_INDEX(rscreen, meq);
+	case R600_QUERY_GPU_ME_BUSY:
+		return BUSY_INDEX(rscreen, me);
+	case R600_QUERY_GPU_SURF_SYNC_BUSY:
+		return BUSY_INDEX(rscreen, surf_sync);
+	case R600_QUERY_GPU_DMA_BUSY:
+		return BUSY_INDEX(rscreen, dma);
+	case R600_QUERY_GPU_SCRATCH_RAM_BUSY:
+		return BUSY_INDEX(rscreen, scratch_ram);
+	case R600_QUERY_GPU_CE_BUSY:
+		return BUSY_INDEX(rscreen, ce);
 	default:
 		unreachable("invalid query type");
 	}
diff --git a/src/gallium/drivers/radeon/r600_pipe_common.h b/src/gallium/drivers/radeon/r600_pipe_common.h
index 9118b6bcf84..fb5bf75df15 100644
--- a/src/gallium/drivers/radeon/r600_pipe_common.h
+++ b/src/gallium/drivers/radeon/r600_pipe_common.h
@@ -377,6 +377,15 @@ union r600_mmio_counters {
 
 		/* SRBM_STATUS2 */
 		struct r600_mmio_counter sdma;
+
+		/* CP_STAT */
+		struct r600_mmio_counter pfp;
+		struct r600_mmio_counter meq;
+		struct r600_mmio_counter me;
+		struct r600_mmio_counter surf_sync;
+		struct r600_mmio_counter dma;
+		struct r600_mmio_counter scratch_ram;
+		struct r600_mmio_counter ce;
 	} named;
 	unsigned array[0];
 };
diff --git a/src/gallium/drivers/radeon/r600_query.c b/src/gallium/drivers/radeon/r600_query.c
index 1747026d0dc..05741d33c9f 100644
--- a/src/gallium/drivers/radeon/r600_query.c
+++ b/src/gallium/drivers/radeon/r600_query.c
@@ -163,6 +163,13 @@ static bool r600_query_sw_begin(struct r600_common_context *rctx,
 	case R600_QUERY_GPU_CP_BUSY:
 	case R600_QUERY_GPU_CB_BUSY:
 	case R600_QUERY_GPU_SDMA_BUSY:
+	case R600_QUERY_GPU_PFP_BUSY:
+	case R600_QUERY_GPU_MEQ_BUSY:
+	case R600_QUERY_GPU_ME_BUSY:
+	case R600_QUERY_GPU_SURF_SYNC_BUSY:
+	case R600_QUERY_GPU_DMA_BUSY:
+	case R600_QUERY_GPU_SCRATCH_RAM_BUSY:
+	case R600_QUERY_GPU_CE_BUSY:
 		query->begin_result = r600_begin_counter(rctx->screen,
 							 query->b.type);
 		break;
@@ -271,6 +278,13 @@ static bool r600_query_sw_end(struct r600_common_context *rctx,
 	case R600_QUERY_GPU_CP_BUSY:
 	case R600_QUERY_GPU_CB_BUSY:
 	case R600_QUERY_GPU_SDMA_BUSY:
+	case R600_QUERY_GPU_PFP_BUSY:
+	case R600_QUERY_GPU_MEQ_BUSY:
+	case R600_QUERY_GPU_ME_BUSY:
+	case R600_QUERY_GPU_SURF_SYNC_BUSY:
+	case R600_QUERY_GPU_DMA_BUSY:
+	case R600_QUERY_GPU_SCRATCH_RAM_BUSY:
+	case R600_QUERY_GPU_CE_BUSY:
 		query->end_result = r600_end_counter(rctx->screen,
 						     query->b.type,
 						     query->begin_result);
@@ -1771,6 +1785,13 @@ static struct pipe_driver_query_info r600_driver_query_list[] = {
 	X("GPU-cp-busy",		GPU_CP_BUSY,		UINT64, AVERAGE),
 	X("GPU-cb-busy",		GPU_CB_BUSY,		UINT64, AVERAGE),
 	X("GPU-sdma-busy",		GPU_SDMA_BUSY,		UINT64, AVERAGE),
+	X("GPU-pfp-busy",		GPU_PFP_BUSY,		UINT64, AVERAGE),
+	X("GPU-meq-busy",		GPU_MEQ_BUSY,		UINT64, AVERAGE),
+	X("GPU-me-busy",		GPU_ME_BUSY,		UINT64, AVERAGE),
+	X("GPU-surf-sync-busy",		GPU_SURF_SYNC_BUSY,	UINT64, AVERAGE),
+	X("GPU-dma-busy",		GPU_DMA_BUSY,		UINT64, AVERAGE),
+	X("GPU-scratch-ram-busy",	GPU_SCRATCH_RAM_BUSY,	UINT64, AVERAGE),
+	X("GPU-ce-busy",		GPU_CE_BUSY,		UINT64, AVERAGE),
 
 	X("temperature",		GPU_TEMPERATURE,	UINT64, AVERAGE),
 	X("shader-clock",		CURRENT_GPU_SCLK,	HZ, AVERAGE),
@@ -1785,10 +1806,14 @@ static unsigned r600_get_num_queries(struct r600_common_screen *rscreen)
 {
 	if (rscreen->info.drm_major == 2 && rscreen->info.drm_minor >= 42)
 		return ARRAY_SIZE(r600_driver_query_list);
-	else if (rscreen->info.drm_major == 3)
-		return ARRAY_SIZE(r600_driver_query_list) - 3;
+	else if (rscreen->info.drm_major == 3) {
+		if (rscreen->chip_class >= VI)
+			return ARRAY_SIZE(r600_driver_query_list) - 3;
+		else
+			return ARRAY_SIZE(r600_driver_query_list) - 10;
+	}
 	else
-		return ARRAY_SIZE(r600_driver_query_list) - 18;
+		return ARRAY_SIZE(r600_driver_query_list) - 25;
 }
 
 static int r600_get_driver_query_info(struct pipe_screen *screen,
diff --git a/src/gallium/drivers/radeon/r600_query.h b/src/gallium/drivers/radeon/r600_query.h
index 0b32793c657..5de80d966c7 100644
--- a/src/gallium/drivers/radeon/r600_query.h
+++ b/src/gallium/drivers/radeon/r600_query.h
@@ -86,6 +86,13 @@ enum {
 	R600_QUERY_GPU_CP_BUSY,
 	R600_QUERY_GPU_CB_BUSY,
 	R600_QUERY_GPU_SDMA_BUSY,
+	R600_QUERY_GPU_PFP_BUSY,
+	R600_QUERY_GPU_MEQ_BUSY,
+	R600_QUERY_GPU_ME_BUSY,
+	R600_QUERY_GPU_SURF_SYNC_BUSY,
+	R600_QUERY_GPU_DMA_BUSY,
+	R600_QUERY_GPU_SCRATCH_RAM_BUSY,
+	R600_QUERY_GPU_CE_BUSY,
 	R600_QUERY_NUM_COMPILATIONS,
 	R600_QUERY_NUM_SHADERS_CREATED,
 	R600_QUERY_BACK_BUFFER_PS_DRAW_RATIO,