Re: [Piglit] [PATCH 5/8] arb_shader_image_load_store: add additional coherency test

2016-04-16 Thread Francisco Jerez
Nicolai Hähnle writes: > On 16.04.2016 16:01, Francisco Jerez wrote: >> Jan Vesely writes: >> >>> On Sat, 2016-04-16 at 10:19 -0500, Nicolai Hähnle wrote: On 15.04.2016 17:12, Francisco Jerez wrote: > >> >>> > > For a test doing almost the same thing but

Re: [Piglit] [PATCH 5/8] arb_shader_image_load_store: add additional coherency test

2016-04-16 Thread Francisco Jerez
Jan Vesely writes: > On Sat, 2016-04-16 at 10:19 -0500, Nicolai Hähnle wrote: >> On 15.04.2016 17:12, Francisco Jerez wrote: >> > >> > > >> > > > >> > > > > >> > > > > > >> > > > > > For a test doing almost the same thing but not relying on >> > > > > > unspecified >> > > > > > invocation or

Re: [Piglit] [PATCH 5/8] arb_shader_image_load_store: add additional coherency test

2016-04-16 Thread Jan Vesely
On Sat, 2016-04-16 at 10:19 -0500, Nicolai Hähnle wrote: > On 15.04.2016 17:12, Francisco Jerez wrote: > > > > > > > > > > > > > > > > > > > > > > > > > > For a test doing almost the same thing but not relying on > > > > > > unspecified > > > > > > invocation ordering, see > > > > > > "tests/s

Re: [Piglit] [PATCH 5/8] arb_shader_image_load_store: add additional coherency test

2016-04-16 Thread Francisco Jerez
Nicolai Hähnle writes: > On 15.04.2016 17:12, Francisco Jerez wrote: >> For a test doing almost the same thing but not relying on unspecified >> invocation ordering, see >> "tests/spec/arb_shader_image_load_store/shader-mem-barrier.c" -- It >> would be interesting to see whether y

Re: [Piglit] [PATCH 5/8] arb_shader_image_load_store: add additional coherency test

2016-04-16 Thread Nicolai Hähnle
On 15.04.2016 17:12, Francisco Jerez wrote: For a test doing almost the same thing but not relying on unspecified invocation ordering, see "tests/spec/arb_shader_image_load_store/shader-mem-barrier.c" -- It would be interesting to see whether you can get it to reproduce the GCN coherency bug usin

Re: [Piglit] [PATCH 5/8] arb_shader_image_load_store: add additional coherency test

2016-04-15 Thread Francisco Jerez
Nicolai Hähnle writes: > On 14.04.2016 23:43, Francisco Jerez wrote: >> Nicolai Hähnle writes: >> >>> On 14.04.2016 22:37, Francisco Jerez wrote: This test seems bogus, ARB_shader_image_load_store doesn't give you any ordering or concurrency guarantees for independent shader invocation

Re: [Piglit] [PATCH 5/8] arb_shader_image_load_store: add additional coherency test

2016-04-15 Thread Nicolai Hähnle
On 14.04.2016 23:43, Francisco Jerez wrote: Nicolai Hähnle writes: On 14.04.2016 22:37, Francisco Jerez wrote: This test seems bogus, ARB_shader_image_load_store doesn't give you any ordering or concurrency guarantees for independent shader invocations (i.e. invocations so that there is no da

Re: [Piglit] [PATCH 5/8] arb_shader_image_load_store: add additional coherency test

2016-04-14 Thread Francisco Jerez
Nicolai Hähnle writes: > On 14.04.2016 22:37, Francisco Jerez wrote: >> This test seems bogus, ARB_shader_image_load_store doesn't give you any >> ordering or concurrency guarantees for independent shader invocations >> (i.e. invocations so that there is no data dependency between the inputs >> o

Re: [Piglit] [PATCH 5/8] arb_shader_image_load_store: add additional coherency test

2016-04-14 Thread Francisco Jerez
This test seems bogus, ARB_shader_image_load_store doesn't give you any ordering or concurrency guarantees for independent shader invocations (i.e. invocations so that there is no data dependency between the inputs of one and the outputs of another). The writer thread may seem to make no progress

[Piglit] [PATCH 5/8] arb_shader_image_load_store: add additional coherency test

2016-04-06 Thread Nicolai Hähnle
From: Nicolai Hähnle The existing coherency test isn't a good match for the AMD GCN execution model. --- .../execution/coherency-extra.shader_test | 90 ++ 1 file changed, 90 insertions(+) create mode 100644 tests/spec/arb_shader_image_load_store/execution/coherenc