Mesa (main): radeonsi/sqtt: setup spm bo

[GitLab Mirror]

Module: Mesa
Branch: main
Commit: e5e8b56236cd287b0cd70f02a64161a2b2031de2
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=e5e8b56236cd287b0cd70f02a64161a2b2031de2

Author: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer at amd.com>
Date:   Tue Mar 29 14:36:26 2022 +0200

radeonsi/sqtt: setup spm bo

The hardware will write SPM counters to this bo.

Reviewed-by: Marek Olšák <marek.olsak at amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15646>

---

 src/gallium/drivers/radeonsi/si_perfcounter.c | 26 ++++++++++++++++++++++++++
 src/gallium/drivers/radeonsi/si_sqtt.c        | 11 ++++++++++-
 2 files changed, 36 insertions(+), 1 deletion(-)

diff --git a/src/gallium/drivers/radeonsi/si_perfcounter.c b/src/gallium/drivers/radeonsi/si_perfcounter.c
index 571424f7fad..d191c44acb8 100644
--- a/src/gallium/drivers/radeonsi/si_perfcounter.c
+++ b/src/gallium/drivers/radeonsi/si_perfcounter.c
@@ -713,6 +713,26 @@ void si_init_perfcounters(struct si_screen *screen)
    }
 }
 
+static bool
+si_spm_init_bo(struct si_context *sctx)
+{
+   struct radeon_winsys *ws = sctx->ws;
+   uint64_t size = 32 * 1024 * 1024; /* Default to 32MB. */
+
+   sctx->spm_trace.buffer_size = size;
+   sctx->spm_trace.sample_interval = 4096; /* Default to 4096 clk. */
+
+   sctx->spm_trace.bo = ws->buffer_create(
+      ws, size, 4096,
+      RADEON_DOMAIN_VRAM,
+      RADEON_FLAG_NO_INTERPROCESS_SHARING |
+         RADEON_FLAG_GTT_WC |
+         RADEON_FLAG_NO_SUBALLOC);
+
+   return sctx->spm_trace.bo != NULL;
+}
+
+
 static void
 si_emit_spm_counters(struct si_context *sctx, struct radeon_cmdbuf *cs)
 {
@@ -886,11 +906,17 @@ si_spm_init(struct si_context *sctx)
    if (!ac_init_spm(info, pc, ARRAY_SIZE(spm_counters), spm_counters, &sctx->spm_trace))
       return false;
 
+   if (!si_spm_init_bo(sctx))
+      return false;
+
    return true;
 }
 
 void
 si_spm_finish(struct si_context *sctx)
 {
+   struct pb_buffer *bo = sctx->spm_trace.bo;
+   radeon_bo_reference(sctx->screen->ws, &bo, NULL);
+
    ac_destroy_spm(&sctx->spm_trace);
 }
diff --git a/src/gallium/drivers/radeonsi/si_sqtt.c b/src/gallium/drivers/radeonsi/si_sqtt.c
index cefbbec0518..a468a5fa1b6 100644
--- a/src/gallium/drivers/radeonsi/si_sqtt.c
+++ b/src/gallium/drivers/radeonsi/si_sqtt.c
@@ -389,6 +389,10 @@ si_thread_trace_start(struct si_context *sctx, int family, struct radeon_cmdbuf
                      sctx->thread_trace->bo,
                      RADEON_USAGE_READWRITE,
                      RADEON_DOMAIN_VRAM);
+   ws->cs_add_buffer(cs,
+                     sctx->spm_trace.bo,
+                     RADEON_USAGE_READWRITE,
+                     RADEON_DOMAIN_VRAM);
 
    si_cp_dma_wait_for_idle(sctx, cs);
 
@@ -432,6 +436,11 @@ si_thread_trace_stop(struct si_context *sctx, int family, struct radeon_cmdbuf *
                      RADEON_USAGE_READWRITE,
                      RADEON_DOMAIN_VRAM);
 
+   ws->cs_add_buffer(cs,
+                     sctx->spm_trace.bo,
+                     RADEON_USAGE_READWRITE,
+                     RADEON_DOMAIN_VRAM);
+
    si_cp_dma_wait_for_idle(sctx, cs);
 
    /* Make sure to wait-for-idle before stopping SQTT. */
@@ -711,7 +720,7 @@ si_handle_thread_trace(struct si_context *sctx, struct radeon_cmdbuf *rcs)
       /* Wait for SQTT to finish and read back the bo */
       if (sctx->ws->fence_wait(sctx->ws, sctx->last_sqtt_fence, PIPE_TIMEOUT_INFINITE) &&
           si_get_thread_trace(sctx, &thread_trace)) {
-         ac_dump_rgp_capture(&sctx->screen->info, &thread_trace, NULL);
+         ac_dump_rgp_capture(&sctx->screen->info, &thread_trace, &sctx->spm_trace);
       } else {
          fprintf(stderr, "Failed to read the trace\n");
       }