radeonsi: print CE IBs into ddebug reports

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

src/gallium/drivers/r600/r600_hw_context.c

@@ -284,7 +284,7 @@ void r600_context_gfx_flush(void *context, unsigned flags,
 	if (ctx->is_debug) {
 		/* Save the IB for debug contexts. */
 		radeon_clear_saved_cs(&ctx->last_gfx);
-		radeon_save_cs(ws, cs, &ctx->last_gfx);
+		radeon_save_cs(ws, cs, &ctx->last_gfx, true);
 		r600_resource_reference(&ctx->last_trace_buf, ctx->trace_buf);
 		r600_resource_reference(&ctx->trace_buf, NULL);
 	}

src/gallium/drivers/radeon/r600_pipe_common.c

@@ -442,7 +442,7 @@ static void r600_flush_dma_ring(void *ctx, unsigned flags,
 	}
 
 	if (check_vm)
-		radeon_save_cs(rctx->ws, cs, &saved);
+		radeon_save_cs(rctx->ws, cs, &saved, true);
 
 	rctx->ws->cs_flush(cs, flags, &rctx->last_sdma_fence);
 	if (fence)
@@ -464,7 +464,7 @@ static void r600_flush_dma_ring(void *ctx, unsigned flags,
  * list in \p saved.
  */
 void radeon_save_cs(struct radeon_winsys *ws, struct radeon_winsys_cs *cs,
-		    struct radeon_saved_cs *saved)
+		    struct radeon_saved_cs *saved, bool get_buffer_list)
 {
 	void *buf;
 	unsigned i;
@@ -482,6 +482,9 @@ void radeon_save_cs(struct radeon_winsys *ws, struct radeon_winsys_cs *cs,
 	}
 	memcpy(buf, cs->current.buf, cs->current.cdw * 4);
 
+	if (!get_buffer_list)
+		return;
+
 	/* Save the buffer list. */
 	saved->bo_count = ws->cs_get_buffer_list(cs, NULL);
 	saved->bo_list = CALLOC(saved->bo_count,

src/gallium/drivers/radeon/r600_pipe_common.h

@@ -767,7 +767,7 @@ const char *r600_get_llvm_processor_name(enum radeon_family family);
 void r600_need_dma_space(struct r600_common_context *ctx, unsigned num_dw,
 			 struct r600_resource *dst, struct r600_resource *src);
 void radeon_save_cs(struct radeon_winsys *ws, struct radeon_winsys_cs *cs,
-		    struct radeon_saved_cs *saved);
+		    struct radeon_saved_cs *saved, bool get_buffer_list);
 void radeon_clear_saved_cs(struct radeon_saved_cs *saved);
 bool r600_check_device_reset(struct r600_common_context *rctx);
 

src/gallium/drivers/radeonsi/si_debug.c

@@ -220,6 +220,7 @@ static void si_dump_debug_registers(struct si_context *sctx, FILE *f)
 static void si_dump_last_ib(struct si_context *sctx, FILE *f)
 {
 	int last_trace_id = -1;
+	int last_ce_trace_id = -1;
 
 	if (!sctx->last_gfx.ib)
 		return;
@@ -233,8 +234,10 @@ static void si_dump_last_ib(struct si_context *sctx, FILE *f)
 						       NULL,
 						       PIPE_TRANSFER_UNSYNCHRONIZED |
 						       PIPE_TRANSFER_READ);
-		if (map)
-			last_trace_id = *map;
+		if (map) {
+			last_trace_id = map[0];
+			last_ce_trace_id = map[1];
+		}
 	}
 
 	if (sctx->init_config)
@@ -251,6 +254,12 @@ static void si_dump_last_ib(struct si_context *sctx, FILE *f)
 	ac_parse_ib(f, sctx->last_gfx.ib, sctx->last_gfx.num_dw,
 		    last_trace_id, "IB", sctx->b.chip_class,
 		     NULL, NULL);
+
+	if (sctx->last_ce.ib) {
+		ac_parse_ib(f, sctx->last_ce.ib, sctx->last_ce.num_dw,
+			    last_ce_trace_id, "CE IB", sctx->b.chip_class,
+			    NULL, NULL);
+	}
 }
 
 static const char *priority_to_string(enum radeon_bo_priority priority)
@@ -845,6 +854,7 @@ static void si_dump_debug_state(struct pipe_context *ctx, FILE *f,
 
 		/* dump only once */
 		radeon_clear_saved_cs(&sctx->last_gfx);
+		radeon_clear_saved_cs(&sctx->last_ce);
 		r600_resource_reference(&sctx->last_trace_buf, NULL);
 	}
 }

src/gallium/drivers/radeonsi/si_hw_context.c

@@ -145,7 +145,9 @@ void si_context_gfx_flush(void *context, unsigned flags,
 	if (ctx->is_debug) {
 		/* Save the IB for debug contexts. */
 		radeon_clear_saved_cs(&ctx->last_gfx);
-		radeon_save_cs(ws, cs, &ctx->last_gfx);
+		radeon_save_cs(ws, cs, &ctx->last_gfx, true);
+		radeon_clear_saved_cs(&ctx->last_ce);
+		radeon_save_cs(ws, ctx->ce_ib, &ctx->last_ce, false);
 		r600_resource_reference(&ctx->last_trace_buf, ctx->trace_buf);
 		r600_resource_reference(&ctx->trace_buf, NULL);
 	}
@@ -173,16 +175,16 @@ void si_context_gfx_flush(void *context, unsigned flags,
 void si_begin_new_cs(struct si_context *ctx)
 {
 	if (ctx->is_debug) {
-		uint32_t zero = 0;
+		static const uint32_t zeros[2];
 
 		/* Create a buffer used for writing trace IDs and initialize it to 0. */
 		assert(!ctx->trace_buf);
 		ctx->trace_buf = (struct r600_resource*)
 				 pipe_buffer_create(ctx->b.b.screen, 0,
-						    PIPE_USAGE_STAGING, 4);
+						    PIPE_USAGE_STAGING, 8);
 		if (ctx->trace_buf)
 			pipe_buffer_write_nooverlap(&ctx->b.b, &ctx->trace_buf->b.b,
-						    0, sizeof(zero), &zero);
+						    0, sizeof(zeros), zeros);
 		ctx->trace_id = 0;
 	}
 

src/gallium/drivers/radeonsi/si_pipe.h

@@ -411,6 +411,7 @@ struct si_context {
 	/* Debug state. */
 	bool			is_debug;
 	struct radeon_saved_cs	last_gfx;
+	struct radeon_saved_cs	last_ce;
 	struct r600_resource	*last_trace_buf;
 	struct r600_resource	*trace_buf;
 	unsigned		trace_id;

src/gallium/drivers/radeonsi/si_state_draw.c

@@ -1409,6 +1409,7 @@ void si_trace_emit(struct si_context *sctx)
 	sctx->trace_id++;
 	radeon_add_to_buffer_list(&sctx->b, &sctx->b.gfx, sctx->trace_buf,
 			      RADEON_USAGE_READWRITE, RADEON_PRIO_TRACE);
+
 	radeon_emit(cs, PKT3(PKT3_WRITE_DATA, 3, 0));
 	radeon_emit(cs, S_370_DST_SEL(V_370_MEMORY_SYNC) |
 		    S_370_WR_CONFIRM(1) |
@@ -1418,4 +1419,18 @@ void si_trace_emit(struct si_context *sctx)
 	radeon_emit(cs, sctx->trace_id);
 	radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
 	radeon_emit(cs, AC_ENCODE_TRACE_POINT(sctx->trace_id));
+
+	if (sctx->ce_ib) {
+		struct radeon_winsys_cs *ce = sctx->ce_ib;
+
+		radeon_emit(ce, PKT3(PKT3_WRITE_DATA, 3, 0));
+		radeon_emit(ce, S_370_DST_SEL(V_370_MEM_ASYNC) |
+			    S_370_WR_CONFIRM(1) |
+			    S_370_ENGINE_SEL(V_370_CE));
+		radeon_emit(ce, sctx->trace_buf->gpu_address + 4);
+		radeon_emit(ce, (sctx->trace_buf->gpu_address + 4) >> 32);
+		radeon_emit(ce, sctx->trace_id);
+		radeon_emit(ce, PKT3(PKT3_NOP, 0, 0));
+		radeon_emit(ce, AC_ENCODE_TRACE_POINT(sctx->trace_id));
+	}
 }