On Wed, Jan 29, 2014 at 04:17:37PM +0000, Damien Lespiau wrote:
> +static void run_test(data_t *data, enum ring r1, enum ring r2, enum test test)
> +{
> + struct ring_ops *r1_ops = &ops[r1];
> + struct ring_ops *r2_ops = &ops[r2];
> + drm_intel_bo *a, *b, *c;
> +
> + a = bo_create(data, WIDTH, HEIGHT, 0xa);
> + b = bo_create(data, WIDTH, HEIGHT, 0xb);
> + c = bo_create(data, WIDTH, HEIGHT, 0xc);
> +
> + r1_ops->busy(data);
> + r2_ops->copy(data, a, b);
^^
r1
> +
> + switch (test) {
> + case TEST_WRITE_READ:
> + r2_ops->copy(data, b, c);
^^
> + bo_check(data, c, 0xa);
> + break;
Turns out I was copying a to b on r2, and then b to c still on r2. So
no inter-ring dependencies in that workload. Ooops.
Pushed an updated patch that copy a to b on r1 and then b to c on r2 and
verified that short-circuiting the semaphores breaks the test indeed.
--
Damien
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
