radeonsi: add HUD queries for cache flush stats

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

diff --git a/src/gallium/drivers/radeon/r600_pipe_common.h b/src/gallium/drivers/radeon/r600_pipe_common.h
index 2e0655602c2123e70ac1e51c3ef804e0ebb06f8a..74f86dc3945682da85ed7547482206ec2c2b5818 100644 (file)
--- a/src/gallium/drivers/radeon/r600_pipe_common.h
+++ b/src/gallium/drivers/radeon/r600_pipe_common.h
@@ -574,6 +574,9 @@ struct r600_common_context {
        unsigned                        num_vs_flushes;
        unsigned                        num_ps_flushes;
        unsigned                        num_cs_flushes;
+       unsigned                        num_fb_cache_flushes;
+       unsigned                        num_L2_invalidates;
+       unsigned                        num_L2_writebacks;
        uint64_t                        num_alloc_tex_transfer_bytes;
        unsigned                        last_tex_ps_draw_ratio; /* for query */
 

diff --git a/src/gallium/drivers/radeon/r600_query.c b/src/gallium/drivers/radeon/r600_query.c
index 0e4270a58cf1beab2e2d21c372360cff61fbe1a8..6b93329e0eeaa30230f9e21a4cea2476f5d2031c 100644 (file)
--- a/src/gallium/drivers/radeon/r600_query.c
+++ b/src/gallium/drivers/radeon/r600_query.c
@@ -113,6 +113,15 @@ static bool r600_query_sw_begin(struct r600_common_context *rctx,
        case R600_QUERY_NUM_CS_FLUSHES:
                query->begin_result = rctx->num_cs_flushes;
                break;
+       case R600_QUERY_NUM_FB_CACHE_FLUSHES:
+               query->begin_result = rctx->num_fb_cache_flushes;
+               break;
+       case R600_QUERY_NUM_L2_INVALIDATES:
+               query->begin_result = rctx->num_L2_invalidates;
+               break;
+       case R600_QUERY_NUM_L2_WRITEBACKS:
+               query->begin_result = rctx->num_L2_writebacks;
+               break;
        case R600_QUERY_REQUESTED_VRAM:
        case R600_QUERY_REQUESTED_GTT:
        case R600_QUERY_MAPPED_VRAM:
@@ -197,6 +206,15 @@ static bool r600_query_sw_end(struct r600_common_context *rctx,
        case R600_QUERY_NUM_CS_FLUSHES:
                query->end_result = rctx->num_cs_flushes;
                break;
+       case R600_QUERY_NUM_FB_CACHE_FLUSHES:
+               query->end_result = rctx->num_fb_cache_flushes;
+               break;
+       case R600_QUERY_NUM_L2_INVALIDATES:
+               query->end_result = rctx->num_L2_invalidates;
+               break;
+       case R600_QUERY_NUM_L2_WRITEBACKS:
+               query->end_result = rctx->num_L2_writebacks;
+               break;
        case R600_QUERY_REQUESTED_VRAM:
        case R600_QUERY_REQUESTED_GTT:
        case R600_QUERY_MAPPED_VRAM:
@@ -1665,6 +1683,9 @@ static struct pipe_driver_query_info r600_driver_query_list[] = {
        X("num-vs-flushes",             NUM_VS_FLUSHES,         UINT64, AVERAGE),
        X("num-ps-flushes",             NUM_PS_FLUSHES,         UINT64, AVERAGE),
        X("num-cs-flushes",             NUM_CS_FLUSHES,         UINT64, AVERAGE),
+       X("num-fb-cache-flushes",       NUM_FB_CACHE_FLUSHES,   UINT64, AVERAGE),
+       X("num-L2-invalidates",         NUM_L2_INVALIDATES,     UINT64, AVERAGE),
+       X("num-L2-writebacks",          NUM_L2_WRITEBACKS,      UINT64, AVERAGE),
        X("requested-VRAM",             REQUESTED_VRAM,         BYTES, AVERAGE),
        X("requested-GTT",              REQUESTED_GTT,          BYTES, AVERAGE),
        X("mapped-VRAM",                MAPPED_VRAM,            BYTES, AVERAGE),

diff --git a/src/gallium/drivers/radeon/r600_query.h b/src/gallium/drivers/radeon/r600_query.h
index 2ff586ac11d936ce3f81bb1ce75e24ef60bbd536..af434faf50bde639105887b91f754d116030b3c1 100644 (file)
--- a/src/gallium/drivers/radeon/r600_query.h
+++ b/src/gallium/drivers/radeon/r600_query.h
@@ -52,6 +52,9 @@ enum {
        R600_QUERY_NUM_VS_FLUSHES,
        R600_QUERY_NUM_PS_FLUSHES,
        R600_QUERY_NUM_CS_FLUSHES,
+       R600_QUERY_NUM_FB_CACHE_FLUSHES,
+       R600_QUERY_NUM_L2_INVALIDATES,
+       R600_QUERY_NUM_L2_WRITEBACKS,
        R600_QUERY_REQUESTED_VRAM,
        R600_QUERY_REQUESTED_GTT,
        R600_QUERY_MAPPED_VRAM,

diff --git a/src/gallium/drivers/radeonsi/si_state_draw.c b/src/gallium/drivers/radeonsi/si_state_draw.c
index cae19dc5a7a37f6ded115dd08b7da500620e8c85..b3f664eff2fb8b93101924494c867ec082f930eb 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_state_draw.c
+++ b/src/gallium/drivers/radeonsi/si_state_draw.c
@@ -735,6 +735,9 @@ void si_emit_cache_flush(struct si_context *sctx)
        struct radeon_winsys_cs *cs = rctx->gfx.cs;
        uint32_t cp_coher_cntl = 0;
 
+       if (rctx->flags & SI_CONTEXT_FLUSH_AND_INV_FRAMEBUFFER)
+               sctx->b.num_fb_cache_flushes++;
+
        /* SI has a bug that it always flushes ICACHE and KCACHE if either
         * bit is set. An alternative way is to write SQC_CACHES, but that
         * doesn't seem to work reliably. Since the bug doesn't affect
@@ -852,6 +855,7 @@ void si_emit_cache_flush(struct si_context *sctx)
                                     S_0085F0_TC_ACTION_ENA(1) |
                                     S_0301F0_TC_WB_ACTION_ENA(rctx->chip_class >= VI));
                cp_coher_cntl = 0;
+               sctx->b.num_L2_invalidates++;
        } else {
                /* L1 invalidation and L2 writeback must be done separately,
                 * because both operations can't be done together.
@@ -867,6 +871,7 @@ void si_emit_cache_flush(struct si_context *sctx)
                                             S_0301F0_TC_WB_ACTION_ENA(1) |
                                             S_0301F0_TC_NC_ACTION_ENA(1));
                        cp_coher_cntl = 0;
+                       sctx->b.num_L2_writebacks++;
                }
                if (rctx->flags & SI_CONTEXT_INV_VMEM_L1) {
                        /* Invalidate per-CU VMEM L1. */