X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fmesa%2Fdrivers%2Fdri%2Fi965%2Fbrw_performance_monitor.c;h=f8e50e10fa3a611c4bc6d9944870044513a18e18;hb=0e657b7b55bc7c83c8eb5258cd9522b0e5e581b7;hp=0a123754257c9b0e7a6c3364cd33622691874d4d;hpb=4b35ab9bdb4e663f41ff5c9ae5bbcc650b6093f9;p=mesa.git

diff --git a/src/mesa/drivers/dri/i965/brw_performance_monitor.c b/src/mesa/drivers/dri/i965/brw_performance_monitor.c
index 0a123754257..f8e50e10fa3 100644
--- a/src/mesa/drivers/dri/i965/brw_performance_monitor.c
+++ b/src/mesa/drivers/dri/i965/brw_performance_monitor.c
@@ -574,10 +574,9 @@ monitor_needs_statistics_registers(struct brw_context *brw,
 static void
 snapshot_statistics_registers(struct brw_context *brw,
                               struct brw_perf_monitor_object *monitor,
-                              uint32_t offset_in_bytes)
+                              uint32_t offset)
 {
    struct gl_context *ctx = &brw->ctx;
-   const int offset = offset_in_bytes / sizeof(uint64_t);
    const int group = PIPELINE_STATS_COUNTERS;
    const int num_counters = ctx->PerfMonitor.Groups[group].NumCounters;
 
@@ -590,7 +589,7 @@ snapshot_statistics_registers(struct brw_context *brw,
 
          brw_store_register_mem64(brw, monitor->pipeline_stats_bo,
                                   brw->perfmon.statistics_registers[i],
-                                  offset + i);
+                                  offset + i * sizeof(uint64_t));
       }
    }
 }
@@ -687,12 +686,12 @@ stop_oa_counters(struct brw_context *brw)
  * The amount of batch space it takes to emit an MI_REPORT_PERF_COUNT snapshot,
  * including the required PIPE_CONTROL flushes.
  *
- * Sandybridge is the worst case scenario: brw_emit_mi_flush
- * expands to three PIPE_CONTROLs which are 4 DWords each.  We have to flush
- * before and after MI_REPORT_PERF_COUNT, so multiply by two.  Finally, add
- * the 3 DWords for MI_REPORT_PERF_COUNT itself.
+ * Sandybridge is the worst case scenario: brw_emit_mi_flush expands to four
+ * PIPE_CONTROLs which are 5 DWords each.  We have to flush before and after
+ * MI_REPORT_PERF_COUNT, so multiply by two.  Finally, add the 3 DWords for
+ * MI_REPORT_PERF_COUNT itself.
  */
-#define MI_REPORT_PERF_COUNT_BATCH_DWORDS (2 * (3 * 4) + 3)
+#define MI_REPORT_PERF_COUNT_BATCH_DWORDS (2 * (4 * 5) + 3)
 
 /**
  * Emit an MI_REPORT_PERF_COUNT command packet.
@@ -710,7 +709,7 @@ emit_mi_report_perf_count(struct brw_context *brw,
    /* Make sure the commands to take a snapshot fits in a single batch. */
    intel_batchbuffer_require_space(brw, MI_REPORT_PERF_COUNT_BATCH_DWORDS * 4,
                                    RENDER_RING);
-   int batch_used = brw->batch.used;
+   int batch_used = USED_BATCH(brw->batch);
 
    /* Reports apparently don't always get written unless we flush first. */
    brw_emit_mi_flush(brw);
@@ -754,7 +753,7 @@ emit_mi_report_perf_count(struct brw_context *brw,
    brw_emit_mi_flush(brw);
 
    (void) batch_used;
-   assert(brw->batch.used - batch_used <= MI_REPORT_PERF_COUNT_BATCH_DWORDS * 4);
+   assert(USED_BATCH(brw->batch) - batch_used <= MI_REPORT_PERF_COUNT_BATCH_DWORDS * 4);
 }
 
 /**
@@ -1017,7 +1016,7 @@ wrap_bookend_bo(struct brw_context *brw)
 }
 
 /* This is fairly arbitrary; the trade off is memory usage vs. extra overhead
- * from wrapping.  On Gen7, 32768 should be enough for for 128 snapshots before
+ * from wrapping.  On Gen7, 32768 should be enough for 128 snapshots before
  * wrapping (since each is 256 bytes).
  */
 #define BOOKEND_BO_SIZE_BYTES 32768
@@ -1386,7 +1385,7 @@ void
 brw_perf_monitor_new_batch(struct brw_context *brw)
 {
    assert(brw->batch.ring == RENDER_RING);
-   assert(brw->gen < 6 || brw->batch.used == 0);
+   assert(brw->gen < 6 || USED_BATCH(brw->batch) == 0);
 
    if (brw->perfmon.oa_users == 0)
       return;