nvc0/ir: always use CG mode for loads from atomic-only buffers
authorIlia Mirkin <imirkin@alum.mit.edu>
Sun, 3 Feb 2019 15:06:24 +0000 (10:06 -0500)
committerIlia Mirkin <imirkin@alum.mit.edu>
Thu, 7 Feb 2019 00:35:57 +0000 (19:35 -0500)
commit4443b6ddf2e08d06f3d0457cf20a2e04244cde37
tree5426c74cbf8912315a8e809c3046b16d71df3dc4
parent399215eb7a0517463e5757c598d6cff6ae2301d0
nvc0/ir: always use CG mode for loads from atomic-only buffers

Atomic operations don't update the local cache, which means that we
would have to issue CCTL operations in order to get the updated values.
When we know that a buffer is primarily used for atomic operations, it's
easier to just avoid the caching at that level entirely.

The same issue persists for non-atomic buffers, which will have to be
fixed separately.

Fixes the failing dEQP-GLES31.functional.atomic_counter.* tests.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Cc: 19.0 <mesa-stable@lists.freedesktop.org>
src/gallium/drivers/nouveau/codegen/nv50_ir_from_tgsi.cpp