On 04/01/2016 07:44 AM, Ilia Mirkin wrote:
On Mar 31, 2016 12:09 PM, "Samuel Pitoiset" <samuel.pitoi...@gmail.com
<mailto:samuel.pitoi...@gmail.com>> wrote:
>
> For Maxwell, the ATOMS instruction can be used to perform atomic
> operations on shared memory instead of this load/store lowering pass.
>
> Signed-off-by: Samuel Pitoiset <samuel.pitoi...@gmail.com
<mailto:samuel.pitoi...@gmail.com>>
> ---
> .../nouveau/codegen/nv50_ir_lowering_nvc0.cpp | 20
+++++++++++++-------
> 1 file changed, 13 insertions(+), 7 deletions(-)
>
> 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 7e6c87a..83e753e 100644
> --- a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
> +++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
> @@ -1294,10 +1294,14 @@ NVC0LoweringPass::handleATOM(Instruction *atom)
> sv = SV_LBASE;
> break;
> case FILE_MEMORY_SHARED:
> - if (targ->getChipset() >= NVISA_GK104_CHIPSET) {
> - handleSharedATOMNVE4(atom);
> - } else {
> - handleSharedATOM(atom);
> + if (targ->getChipset() < NVISA_GM107_CHIPSET) {
> + // For Fermi/Kepler, we have to use ld lock/st unlock to
perform
> + // atomic operations on shared memory. For Maxwell, ATOMS
is enough.
> + if (targ->getChipset() >= NVISA_GK104_CHIPSET) {
> + handleSharedATOMNVE4(atom);
> + } else {
> + handleSharedATOM(atom);
> + }
> }
Or, with fewer branches...
If (<nve4) do nvc0 thing
Else if (<gm107) do nve4 thing
Which will drop the outer if, making it more comprehensible.
Looks better, thanks!
> return true;
> default:
> @@ -1326,9 +1330,11 @@ NVC0LoweringPass::handleATOM(Instruction *atom)
> bool
> NVC0LoweringPass::handleCasExch(Instruction *cas, bool needCctl)
> {
> - if (cas->src(0).getFile() == FILE_MEMORY_SHARED) {
> - // ATOM_CAS and ATOM_EXCH are handled in handleSharedATOM().
> - return false;
> + if (targ->getChipset() < NVISA_GM107_CHIPSET) {
> + if (cas->src(0).getFile() == FILE_MEMORY_SHARED) {
> + // ATOM_CAS and ATOM_EXCH are handled in handleSharedATOM().
> + return false;
> + }
> }
>
> if (cas->subOp != NV50_IR_SUBOP_ATOM_CAS &&
> --
> 2.7.4
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev@lists.freedesktop.org <mailto:mesa-dev@lists.freedesktop.org>
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
--
-Samuel
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
