diff --git a/src/broadcom/compiler/nir_to_vir.c b/src/broadcom/compiler/nir_to_vir.c
index 6ec20f8de08..4759f3b0e6b 100644
--- a/src/broadcom/compiler/nir_to_vir.c
+++ b/src/broadcom/compiler/nir_to_vir.c
@@ -324,6 +324,7 @@ ntq_add_pending_tmu_flush(struct v3d_compile *c,
         c->tmu.flush[c->tmu.flush_count].dest = dest;
         c->tmu.flush[c->tmu.flush_count].component_mask = component_mask;
         c->tmu.flush_count++;
+        c->tmu.total_count++;
 
         if (c->disable_tmu_pipelining)
                 ntq_flush_tmu(c);
diff --git a/src/broadcom/compiler/v3d_compiler.h b/src/broadcom/compiler/v3d_compiler.h
index e803ba222bc..cc23376f514 100644
--- a/src/broadcom/compiler/v3d_compiler.h
+++ b/src/broadcom/compiler/v3d_compiler.h
@@ -645,6 +645,7 @@ struct v3d_compile {
                         uint8_t component_mask;
                 } flush[MAX_TMU_QUEUE_SIZE];
                 uint32_t flush_count;
+                uint32_t total_count;
         } tmu;
 
         /**
@@ -919,6 +920,7 @@ struct v3d_prog_data {
         uint32_t spill_size;
         uint32_t tmu_spills;
         uint32_t tmu_fills;
+        uint32_t tmu_count;
 
         uint32_t qpu_read_stalls;
 
diff --git a/src/broadcom/compiler/vir.c b/src/broadcom/compiler/vir.c
index d253be2935a..02f2aacf3e0 100644
--- a/src/broadcom/compiler/vir.c
+++ b/src/broadcom/compiler/vir.c
@@ -855,6 +855,7 @@ v3d_set_prog_data(struct v3d_compile *c,
         prog_data->spill_size = c->spill_size;
         prog_data->tmu_spills = c->spills;
         prog_data->tmu_fills = c->fills;
+        prog_data->tmu_count = c->tmu.total_count;
         prog_data->qpu_read_stalls = c->qpu_inst_stalled_count;
         prog_data->compile_strategy_idx = c->compile_strategy_idx;
         prog_data->tmu_dirty_rcl = c->tmu_dirty_rcl;