Quoting Matthew Auld (2017-11-22 21:21:39)
> Rather than repeat the test for each engine, which takes a long time,
> let's try alternating between the engines in some randomized
> order.
>
> Suggested-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
> Signed-off-by: Matthew Auld <matthew.auld@xxxxxxxxx>
> Cc: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
> ---
> drivers/gpu/drm/i915/selftests/huge_pages.c | 102 ++++++++++++++++------------
> 1 file changed, 57 insertions(+), 45 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/selftests/huge_pages.c b/drivers/gpu/drm/i915/selftests/huge_pages.c
> index db7a0a1f2960..4809368dfbbc 100644
> --- a/drivers/gpu/drm/i915/selftests/huge_pages.c
> +++ b/drivers/gpu/drm/i915/selftests/huge_pages.c
> @@ -27,6 +27,7 @@
> #include <linux/prime_numbers.h>
>
> #include "mock_drm.h"
> +#include "i915_random.h"
>
> static const unsigned int page_sizes[] = {
> I915_GTT_PAGE_SIZE_2M,
> @@ -1044,7 +1045,10 @@ static int igt_write_huge(struct i915_gem_context *ctx,
> {
> struct drm_i915_private *i915 = to_i915(obj->base.dev);
> struct i915_address_space *vm = ctx->ppgtt ? &ctx->ppgtt->base : &i915->ggtt.base;
> + static struct intel_engine_cs *engines[I915_NUM_ENGINES];
> struct intel_engine_cs *engine;
> + I915_RND_STATE(prng);
> + IGT_TIMEOUT(end_time);
> struct i915_vma *vma;
> unsigned int flags = PIN_USER | PIN_OFFSET_FIXED;
> unsigned int max_page_size;
> @@ -1052,6 +1056,8 @@ static int igt_write_huge(struct i915_gem_context *ctx,
> u64 max;
> u64 num;
> u64 size;
> + int *order;
> + int i, n;
> int err = 0;
>
> GEM_BUG_ON(!i915_gem_object_has_pinned_pages(obj));
> @@ -1067,67 +1073,72 @@ static int igt_write_huge(struct i915_gem_context *ctx,
> if (IS_ERR(vma))
> return PTR_ERR(vma);
>
> + n = 0;
> for_each_engine(engine, i915, id) {
> - IGT_TIMEOUT(end_time);
> -
> if (!intel_engine_can_store_dword(engine)) {
> - pr_info("store-dword-imm not supported on engine=%u\n",
> - id);
> + pr_info("store-dword-imm not supported on engine=%u\n", id);
> continue;
> }
> + engines[n++] = engine;
> + }
>
> - /*
> - * Try various offsets until we timeout -- we want to avoid
> - * issues hidden by effectively always using offset = 0.
> - */
> - for_each_prime_number_from(num, 0, max) {
> - u64 offset = num * max_page_size;
> - u32 dword;
> + GEM_BUG_ON(!n);
Think gen2! if (!n) return 0;
> + order = i915_random_order(n, &prng);
if (!order) return -ENOMEM;
Now, I would be a little more cunning in the construction of the
permutation. Use some_number*n, then engine = order[i] % n;
That way we may test feeding to the same engine a few times in succession,
again exploiting the random pattern.
Overall makes a lot of sense, and does seem to be a good compromise.
-Chris
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
