From: Marek Olšák <marek.ol...@amd.com>
---
src/gallium/drivers/radeon/r600_query.c | 4 ++++
src/gallium/drivers/radeon/r600_query.h | 1 +
src/gallium/drivers/radeon/radeon_winsys.h | 1 +
src/gallium/winsys/amdgpu/drm/amdgpu_cs.c | 6 +++++-
src/gallium/winsys/amdgpu/drm/amdgpu_winsys.c | 2 ++
src/gallium/winsys/amdgpu/drm/amdgpu_winsys.h | 1 +
src/gallium/winsys/radeon/drm/radeon_drm_cs.c | 5 ++++-
src/gallium/winsys/radeon/drm/radeon_drm_winsys.c | 2 ++
src/gallium/winsys/radeon/drm/radeon_drm_winsys.h | 1 +
9 files changed, 21 insertions(+), 2 deletions(-)
diff --git a/src/gallium/drivers/radeon/r600_query.c
b/src/gallium/drivers/radeon/r600_query.c
index 70a2568..3c72f27 100644
--- a/src/gallium/drivers/radeon/r600_query.c
+++ b/src/gallium/drivers/radeon/r600_query.c
@@ -59,20 +59,21 @@ static void r600_query_sw_destroy(struct
r600_common_context *rctx,
static enum radeon_value_id winsys_id_from_type(unsigned type)
{
switch (type) {
case R600_QUERY_REQUESTED_VRAM: return RADEON_REQUESTED_VRAM_MEMORY;
case R600_QUERY_REQUESTED_GTT: return RADEON_REQUESTED_GTT_MEMORY;
case R600_QUERY_MAPPED_VRAM: return RADEON_MAPPED_VRAM;
case R600_QUERY_MAPPED_GTT: return RADEON_MAPPED_GTT;
case R600_QUERY_BUFFER_WAIT_TIME: return RADEON_BUFFER_WAIT_TIME_NS;
case R600_QUERY_NUM_GFX_IBS: return RADEON_NUM_GFX_IBS;
+ case R600_QUERY_NUM_SDMA_IBS: return RADEON_NUM_SDMA_IBS;
case R600_QUERY_NUM_BYTES_MOVED: return RADEON_NUM_BYTES_MOVED;
case R600_QUERY_NUM_EVICTIONS: return RADEON_NUM_EVICTIONS;
case R600_QUERY_VRAM_USAGE: return RADEON_VRAM_USAGE;
case R600_QUERY_GTT_USAGE: return RADEON_GTT_USAGE;
case R600_QUERY_GPU_TEMPERATURE: return RADEON_GPU_TEMPERATURE;
case R600_QUERY_CURRENT_GPU_SCLK: return RADEON_CURRENT_SCLK;
case R600_QUERY_CURRENT_GPU_MCLK: return RADEON_CURRENT_MCLK;
default: unreachable("query type does not correspond to winsys id");
}
}
@@ -129,20 +130,21 @@ static bool r600_query_sw_begin(struct
r600_common_context *rctx,
case R600_QUERY_VRAM_USAGE:
case R600_QUERY_GTT_USAGE:
case R600_QUERY_GPU_TEMPERATURE:
case R600_QUERY_CURRENT_GPU_SCLK:
case R600_QUERY_CURRENT_GPU_MCLK:
case R600_QUERY_BACK_BUFFER_PS_DRAW_RATIO:
query->begin_result = 0;
break;
case R600_QUERY_BUFFER_WAIT_TIME:
case R600_QUERY_NUM_GFX_IBS:
+ case R600_QUERY_NUM_SDMA_IBS:
case R600_QUERY_NUM_BYTES_MOVED:
case R600_QUERY_NUM_EVICTIONS: {
enum radeon_value_id ws_id = winsys_id_from_type(query->b.type);
query->begin_result = rctx->ws->query_value(rctx->ws, ws_id);
break;
}
case R600_QUERY_GPU_LOAD:
query->begin_result = r600_gpu_load_begin(rctx->screen);
break;
case R600_QUERY_NUM_COMPILATIONS:
@@ -219,20 +221,21 @@ static bool r600_query_sw_end(struct r600_common_context
*rctx,
case R600_QUERY_REQUESTED_GTT:
case R600_QUERY_MAPPED_VRAM:
case R600_QUERY_MAPPED_GTT:
case R600_QUERY_VRAM_USAGE:
case R600_QUERY_GTT_USAGE:
case R600_QUERY_GPU_TEMPERATURE:
case R600_QUERY_CURRENT_GPU_SCLK:
case R600_QUERY_CURRENT_GPU_MCLK:
case R600_QUERY_BUFFER_WAIT_TIME:
case R600_QUERY_NUM_GFX_IBS:
+ case R600_QUERY_NUM_SDMA_IBS:
case R600_QUERY_NUM_BYTES_MOVED:
case R600_QUERY_NUM_EVICTIONS: {
enum radeon_value_id ws_id = winsys_id_from_type(query->b.type);
query->end_result = rctx->ws->query_value(rctx->ws, ws_id);
break;
}
case R600_QUERY_GPU_LOAD:
query->end_result = r600_gpu_load_end(rctx->screen,
query->begin_result);
query->begin_result = 0;
@@ -1685,20 +1688,21 @@ static struct pipe_driver_query_info
r600_driver_query_list[] = {
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),
X("mapped-GTT", MAPPED_GTT, BYTES,
AVERAGE),
X("buffer-wait-time", BUFFER_WAIT_TIME, MICROSECONDS,
CUMULATIVE),
X("num-GFX-IBs", NUM_GFX_IBS, UINT64, AVERAGE),
+ X("num-SDMA-IBs", NUM_SDMA_IBS, UINT64, AVERAGE),
X("num-bytes-moved", NUM_BYTES_MOVED, BYTES,
CUMULATIVE),
X("num-evictions", NUM_EVICTIONS, UINT64,
CUMULATIVE),
X("VRAM-usage", VRAM_USAGE, BYTES,
AVERAGE),
X("GTT-usage", GTT_USAGE, BYTES,
AVERAGE),
X("back-buffer-ps-draw-ratio", BACK_BUFFER_PS_DRAW_RATIO,
UINT64, AVERAGE),
/* GPIN queries are for the benefit of old versions of GPUPerfStudio,
* which use it as a fallback path to detect the GPU type.
*
* Note: The names of these queries are significant for GPUPerfStudio
diff --git a/src/gallium/drivers/radeon/r600_query.h
b/src/gallium/drivers/radeon/r600_query.h
index fffd9fe..3791ec6 100644
--- a/src/gallium/drivers/radeon/r600_query.h
+++ b/src/gallium/drivers/radeon/r600_query.h
@@ -54,20 +54,21 @@ enum {
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,
R600_QUERY_MAPPED_GTT,
R600_QUERY_BUFFER_WAIT_TIME,
R600_QUERY_NUM_GFX_IBS,
+ R600_QUERY_NUM_SDMA_IBS,
R600_QUERY_NUM_BYTES_MOVED,
R600_QUERY_NUM_EVICTIONS,
R600_QUERY_VRAM_USAGE,
R600_QUERY_GTT_USAGE,
R600_QUERY_GPU_TEMPERATURE,
R600_QUERY_CURRENT_GPU_SCLK,
R600_QUERY_CURRENT_GPU_MCLK,
R600_QUERY_GPU_LOAD,
R600_QUERY_NUM_COMPILATIONS,
R600_QUERY_NUM_SHADERS_CREATED,
diff --git a/src/gallium/drivers/radeon/radeon_winsys.h
b/src/gallium/drivers/radeon/radeon_winsys.h
index 6c886cf..e6fb2d5 100644
--- a/src/gallium/drivers/radeon/radeon_winsys.h
+++ b/src/gallium/drivers/radeon/radeon_winsys.h
@@ -76,20 +76,21 @@ enum ring_type {
};
enum radeon_value_id {
RADEON_REQUESTED_VRAM_MEMORY,
RADEON_REQUESTED_GTT_MEMORY,
RADEON_MAPPED_VRAM,
RADEON_MAPPED_GTT,
RADEON_BUFFER_WAIT_TIME_NS,
RADEON_TIMESTAMP,
RADEON_NUM_GFX_IBS,
+ RADEON_NUM_SDMA_IBS,
RADEON_NUM_BYTES_MOVED,
RADEON_NUM_EVICTIONS,
RADEON_VRAM_USAGE,
RADEON_GTT_USAGE,
RADEON_GPU_TEMPERATURE, /* DRM 2.42.0 */
RADEON_CURRENT_SCLK,
RADEON_CURRENT_MCLK,
RADEON_GPU_RESET_COUNTER, /* DRM 2.43.0 */
};
diff --git a/src/gallium/winsys/amdgpu/drm/amdgpu_cs.c
b/src/gallium/winsys/amdgpu/drm/amdgpu_cs.c
index adb46c5..95402bf 100644
--- a/src/gallium/winsys/amdgpu/drm/amdgpu_cs.c
+++ b/src/gallium/winsys/amdgpu/drm/amdgpu_cs.c
@@ -1197,21 +1197,25 @@ static int amdgpu_cs_flush(struct radeon_winsys_cs *rcs,
amdgpu_get_new_ib(&ws->base, cs, IB_MAIN);
if (cs->const_ib.ib_mapped)
amdgpu_get_new_ib(&ws->base, cs, IB_CONST);
if (cs->const_preamble_ib.ib_mapped)
amdgpu_get_new_ib(&ws->base, cs, IB_CONST_PREAMBLE);
cs->main.base.used_gart = 0;
cs->main.base.used_vram = 0;
- ws->num_gfx_IBs++;
+ if (cs->ring_type == RING_GFX)
+ ws->num_gfx_IBs++;
+ else if (cs->ring_type == RING_DMA)
+ ws->num_sdma_IBs++;
+
return error_code;
}
static void amdgpu_cs_destroy(struct radeon_winsys_cs *rcs)
{
struct amdgpu_cs *cs = amdgpu_cs(rcs);
amdgpu_cs_sync_flush(rcs);
util_queue_fence_destroy(&cs->flush_completed);
p_atomic_dec(&cs->ctx->ws->num_cs);
diff --git a/src/gallium/winsys/amdgpu/drm/amdgpu_winsys.c
b/src/gallium/winsys/amdgpu/drm/amdgpu_winsys.c
index 2ccc8e5..b950d37 100644
--- a/src/gallium/winsys/amdgpu/drm/amdgpu_winsys.c
+++ b/src/gallium/winsys/amdgpu/drm/amdgpu_winsys.c
@@ -422,20 +422,22 @@ static uint64_t amdgpu_query_value(struct radeon_winsys
*rws,
return ws->mapped_vram;
case RADEON_MAPPED_GTT:
return ws->mapped_gtt;
case RADEON_BUFFER_WAIT_TIME_NS:
return ws->buffer_wait_time;
case RADEON_TIMESTAMP:
amdgpu_query_info(ws->dev, AMDGPU_INFO_TIMESTAMP, 8, &retval);
return retval;
case RADEON_NUM_GFX_IBS:
return ws->num_gfx_IBs;
+ case RADEON_NUM_SDMA_IBS:
+ return ws->num_sdma_IBs;
case RADEON_NUM_BYTES_MOVED:
amdgpu_query_info(ws->dev, AMDGPU_INFO_NUM_BYTES_MOVED, 8, &retval);
return retval;
case RADEON_NUM_EVICTIONS:
amdgpu_query_info(ws->dev, AMDGPU_INFO_NUM_EVICTIONS, 8, &retval);
return retval;
case RADEON_VRAM_USAGE:
amdgpu_query_heap_info(ws->dev, AMDGPU_GEM_DOMAIN_VRAM, 0, &heap);
return heap.heap_usage;
case RADEON_GTT_USAGE:
diff --git a/src/gallium/winsys/amdgpu/drm/amdgpu_winsys.h
b/src/gallium/winsys/amdgpu/drm/amdgpu_winsys.h
index 6746079..2a7900a 100644
--- a/src/gallium/winsys/amdgpu/drm/amdgpu_winsys.h
+++ b/src/gallium/winsys/amdgpu/drm/amdgpu_winsys.h
@@ -55,20 +55,21 @@ struct amdgpu_winsys {
pipe_mutex bo_fence_lock;
int num_cs; /* The number of command streams created. */
uint32_t next_bo_unique_id;
uint64_t allocated_vram;
uint64_t allocated_gtt;
uint64_t mapped_vram;
uint64_t mapped_gtt;
uint64_t buffer_wait_time; /* time spent in buffer_wait in ns */
uint64_t num_gfx_IBs;
+ uint64_t num_sdma_IBs;
struct radeon_info info;
/* multithreaded IB submission */
struct util_queue cs_queue;
struct amdgpu_gpu_info amdinfo;
ADDR_HANDLE addrlib;
uint32_t rev_id;
unsigned family;
diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_cs.c
b/src/gallium/winsys/radeon/drm/radeon_drm_cs.c
index ab1ca06..8f9e8a6 100644
--- a/src/gallium/winsys/radeon/drm/radeon_drm_cs.c
+++ b/src/gallium/winsys/radeon/drm/radeon_drm_cs.c
@@ -684,21 +684,24 @@ static int radeon_drm_cs_flush(struct radeon_winsys_cs
*rcs,
} else {
radeon_cs_context_cleanup(cs->cst);
}
/* Prepare a new CS. */
cs->base.current.buf = cs->csc->buf;
cs->base.current.cdw = 0;
cs->base.used_vram = 0;
cs->base.used_gart = 0;
- cs->ws->num_gfx_IBs++;
+ if (cs->ring_type == RING_GFX)
+ cs->ws->num_gfx_IBs++;
+ else if (cs->ring_type == RING_DMA)
+ cs->ws->num_sdma_IBs++;
return 0;
}
static void radeon_drm_cs_destroy(struct radeon_winsys_cs *rcs)
{
struct radeon_drm_cs *cs = radeon_drm_cs(rcs);
radeon_drm_cs_sync_flush(rcs);
util_queue_fence_destroy(&cs->flush_completed);
radeon_cs_context_cleanup(&cs->csc1);
diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_winsys.c
b/src/gallium/winsys/radeon/drm/radeon_drm_winsys.c
index 1b2f9b1..c85e427 100644
--- a/src/gallium/winsys/radeon/drm/radeon_drm_winsys.c
+++ b/src/gallium/winsys/radeon/drm/radeon_drm_winsys.c
@@ -618,20 +618,22 @@ static uint64_t radeon_query_value(struct radeon_winsys
*rws,
if (ws->info.drm_minor < 20 || ws->gen < DRV_R600) {
assert(0);
return 0;
}
radeon_get_drm_value(ws->fd, RADEON_INFO_TIMESTAMP, "timestamp",
(uint32_t*)&retval);
return retval;
case RADEON_NUM_GFX_IBS:
return ws->num_gfx_IBs;
+ case RADEON_NUM_SDMA_IBS:
+ return ws->num_sdma_IBs;
case RADEON_NUM_BYTES_MOVED:
radeon_get_drm_value(ws->fd, RADEON_INFO_NUM_BYTES_MOVED,
"num-bytes-moved", (uint32_t*)&retval);
return retval;
case RADEON_NUM_EVICTIONS:
return 0; /* unimplemented */
case RADEON_VRAM_USAGE:
radeon_get_drm_value(ws->fd, RADEON_INFO_VRAM_USAGE,
"vram-usage", (uint32_t*)&retval);
return retval;
diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_winsys.h
b/src/gallium/winsys/radeon/drm/radeon_drm_winsys.h
index 137a9f4..ac623a1 100644
--- a/src/gallium/winsys/radeon/drm/radeon_drm_winsys.h
+++ b/src/gallium/winsys/radeon/drm/radeon_drm_winsys.h
@@ -73,20 +73,21 @@ struct radeon_drm_winsys {
struct pb_slabs bo_slabs;
int fd; /* DRM file descriptor */
int num_cs; /* The number of command streams created. */
uint64_t allocated_vram;
uint64_t allocated_gtt;
uint64_t mapped_vram;
uint64_t mapped_gtt;
uint64_t buffer_wait_time; /* time spent in buffer_wait in ns */
uint64_t num_gfx_IBs;
+ uint64_t num_sdma_IBs;
uint32_t next_bo_hash;
enum radeon_generation gen;
struct radeon_info info;
uint32_t va_start;
uint32_t va_unmap_working;
uint32_t accel_working2;
/* List of buffer GEM names. Protected by bo_handles_mutex. */
struct util_hash_table *bo_names;
