cell: start some performance measurements

Use the spu_write_decrementer() and spu_read_decrementer() functions to
measure time.  Convert to milliseconds according to the system timebase value.

2 files changed, 23 insertions, 1 deletions

diff --git a/src/gallium/drivers/cell/spu/spu_command.c b/src/gallium/drivers/cell/spu/spu_command.c
index d2c282a022..57d265fef7 100644
--- a/src/gallium/drivers/cell/spu/spu_command.c
+++ b/src/gallium/drivers/cell/spu/spu_command.c
@@ -670,6 +670,8 @@ cmd_batch(uint opcode)
 }
 
 
+#define PERF 0
+
 
 /**
  * Main loop for SPEs: Get a command, execute it, repeat.
@@ -678,6 +680,7 @@ void
 command_loop(void)
 {
    int exitFlag = 0;
+   uint t0, t1;
 
    D_PRINTF(CELL_DEBUG_CMD, "Enter command loop\n");
 
@@ -686,10 +689,16 @@ command_loop(void)
 
       D_PRINTF(CELL_DEBUG_CMD, "Wait for cmd...\n");
 
+      if (PERF)
+         spu_write_decrementer(~0);
+
       /* read/wait from mailbox */
       opcode = (unsigned int) spu_read_in_mbox();
       D_PRINTF(CELL_DEBUG_CMD, "got cmd 0x%x\n", opcode);
 
+      if (PERF)
+         t0 = spu_read_decrementer();
+
       switch (opcode & CELL_CMD_OPCODE_MASK) {
       case CELL_CMD_EXIT:
          D_PRINTF(CELL_DEBUG_CMD, "EXIT\n");
@@ -707,6 +716,12 @@ command_loop(void)
          printf("Bad opcode 0x%x!\n", opcode & CELL_CMD_OPCODE_MASK);
       }
 
+      if (PERF) {
+         t1 = spu_read_decrementer();
+         printf("wait mbox time: %gms   batch time: %gms\n",
+                (~0u - t0) * spu.init.inv_timebase,
+                (t0 - t1) * spu.init.inv_timebase);
+      }
    }
 
    D_PRINTF(CELL_DEBUG_CMD, "Exit command loop\n");
diff --git a/src/gallium/drivers/cell/spu/spu_render.c b/src/gallium/drivers/cell/spu/spu_render.c
index 802455bf79..5515bb55c9 100644
--- a/src/gallium/drivers/cell/spu/spu_render.c
+++ b/src/gallium/drivers/cell/spu/spu_render.c
@@ -175,6 +175,7 @@ cmd_render(const struct cell_command_render *render, uint *pos_incr)
    const ubyte *vertices;
    const ushort *indexes;
    uint i, j;
+   uint num_tiles;
 
    D_PRINTF(CELL_DEBUG_CMD,
             "RENDER prim=%u num_vert=%u num_ind=%u inline_vert=%u\n",
@@ -242,6 +243,8 @@ cmd_render(const struct cell_command_render *render, uint *pos_incr)
    wait_on_mask(1 << TAG_SURFACE_CLEAR); /* XXX temporary */
 
 
+   num_tiles = 0;
+
    /**
     ** loop over tiles, rendering tris
     **/
@@ -255,6 +258,8 @@ cmd_render(const struct cell_command_render *render, uint *pos_incr)
       if (!my_tile(tx, ty))
          continue;
 
+      num_tiles++;
+
       spu.cur_ctile_status = spu.ctile_status[ty][tx];
       spu.cur_ztile_status = spu.ztile_status[ty][tx];
 
@@ -284,5 +289,7 @@ cmd_render(const struct cell_command_render *render, uint *pos_incr)
       spu.ztile_status[ty][tx] = spu.cur_ztile_status;
    }
 
-   D_PRINTF(CELL_DEBUG_CMD, "RENDER done\n");
+   D_PRINTF(CELL_DEBUG_CMD,
+            "RENDER done (%u tiles hit)\n",
+            num_tiles);
 }