[Mesa-dev] [PATCH v2] nvc0/ir: clamp the UBO index for compute on Kepler
Ilia Mirkin
imirkin at alum.mit.edu
Mon Jun 13 15:29:01 UTC 2016
On Mon, Jun 13, 2016 at 11:13 AM, Samuel Pitoiset
<samuel.pitoiset at gmail.com> wrote:
> We already check that the address is not "too far", but we should also
> clamp the UBO index in order to avoid looking at the wrong place in the
> driver cb. This is a pretty rare situation though.
>
> v2: clamp the file index too
>
> Signed-off-by: Samuel Pitoiset <samuel.pitoiset at gmail.com>
> Cc: "12.0" <mesa-stable at lists.freedesktop.org>
> ---
> src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp | 10 +++++++++-
> 1 file changed, 9 insertions(+), 1 deletion(-)
>
> diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
> index cb82216..52cb438 100644
> --- a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
> +++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
> @@ -2183,7 +2183,15 @@ NVC0LoweringPass::handleLDST(Instruction *i)
> int8_t fileIndex = i->getSrc(0)->reg.fileIndex - 1;
> Value *ind = i->getIndirect(0, 1);
>
> - // TODO: clamp the offset to the maximum number of const buf.
> + if (ind) {
> + // Clamp the UBO index when an indirect access is used to avoid
> + // loading information from the wrong place in the driver cb.
> + ind = bld.mkOp2v(OP_MIN, TYPE_U32, ind,
> + bld.mkOp2v(OP_ADD, TYPE_U32, bld.getSSA(),
> + bld.loadImm(NULL, fileIndex), ind),
Let's do the optimizer a favor and flip the order of these args - the
immediate should always be the second arg. The optimizer will do the
flip optionally, but might as well not make it work for it. With that
minor change,
Reviewed-by: Ilia Mirkin <imirkin at alum.mit.edu>
> + bld.loadImm(NULL, 12));
> + }
> +
> if (i->src(0).isIndirect(1)) {
> Value *offset = bld.loadImm(NULL, i->getSrc(0)->reg.data.offset + typeSizeof(i->sType));
> Value *ptr = loadUboInfo64(ind, fileIndex * 16);
> --
> 2.8.3
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
More information about the mesa-dev
mailing list