[Mesa-dev] [PATCH] spirv/nir: Implement OpAtomicLoad/Store for shared variables

[Jason Ekstrand]

On Wed, Sep 7, 2016 at 9:17 AM, Lionel Landwerlin <
lionel.g.landwerlin at intel.com> wrote:

> Thanks,
>
> I thought about that but the various nir_intrinsic_set_write_mask()/etc..
> for load/store kind of discouraged me.
> I'll give it a try though.
>

I just looked a bit and it's going to be a *major* change.  maybe best left
alone for the moment?


>
> On 07/09/16 17:12, Jason Ekstrand wrote:
>
> LGTM.
>
> Reviewed-by: Jason Ekstrand <jason at jlekstrand.net>
>
> Maybe we should consider making the orders of arguments a bit more
> consistent for a few functions so that we can use
> fill_common_atomic_sources for load and store too.
>
> --Jason
>
> On Wed, Sep 7, 2016 at 3:10 AM, Lionel Landwerlin <llandwerlin at gmail.com>
> wrote:
>
>> Missing bits from 2afb950161f847d9b0a7be03dfb62cacc5ea51ba.
>>
>> Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin at intel.com>
>> Cc: Jason Ekstrand <jason.ekstrand at intel.com>
>> ---
>>  src/compiler/spirv/spirv_to_nir.c | 35 ++++++++++++++++++++++++++++++
>> ++++-
>>  1 file changed, 34 insertions(+), 1 deletion(-)
>>
>> diff --git a/src/compiler/spirv/spirv_to_nir.c
>> b/src/compiler/spirv/spirv_to_nir.c
>> index 4d88fdb..f1cd4be 100644
>> --- a/src/compiler/spirv/spirv_to_nir.c
>> +++ b/src/compiler/spirv/spirv_to_nir.c
>> @@ -1920,11 +1920,44 @@ vtn_handle_ssbo_or_shared_atomic(struct
>> vtn_builder *b, SpvOp opcode,
>>     */
>>
>>     if (chain->var->mode == vtn_variable_mode_workgroup) {
>> +      struct vtn_type *type = chain->var->type;
>>        nir_deref *deref = &vtn_access_chain_to_deref(b, chain)->deref;
>>        nir_intrinsic_op op = get_shared_nir_atomic_op(opcode);
>>        atomic = nir_intrinsic_instr_create(b->nb.shader, op);
>>        atomic->variables[0] = nir_deref_as_var(nir_copy_deref(atomic,
>> deref));
>> -      fill_common_atomic_sources(b, opcode, w, &atomic->src[0]);
>> +
>> +      switch (opcode) {
>> +      case SpvOpAtomicLoad:
>> +         atomic->num_components = glsl_get_vector_elements(type->type);
>> +         break;
>> +
>> +      case SpvOpAtomicStore:
>> +         atomic->num_components = glsl_get_vector_elements(type->type);
>> +         nir_intrinsic_set_write_mask(atomic, (1 <<
>> atomic->num_components) - 1);
>> +         atomic->src[0] = nir_src_for_ssa(vtn_ssa_value(b, w[4])->def);
>> +         break;
>> +
>> +      case SpvOpAtomicExchange:
>> +      case SpvOpAtomicCompareExchange:
>> +      case SpvOpAtomicCompareExchangeWeak:
>> +      case SpvOpAtomicIIncrement:
>> +      case SpvOpAtomicIDecrement:
>> +      case SpvOpAtomicIAdd:
>> +      case SpvOpAtomicISub:
>> +      case SpvOpAtomicSMin:
>> +      case SpvOpAtomicUMin:
>> +      case SpvOpAtomicSMax:
>> +      case SpvOpAtomicUMax:
>> +      case SpvOpAtomicAnd:
>> +      case SpvOpAtomicOr:
>> +      case SpvOpAtomicXor:
>> +         fill_common_atomic_sources(b, opcode, w, &atomic->src[0]);
>> +         break;
>> +
>> +      default:
>> +         unreachable("Invalid SPIR-V atomic");
>> +
>> +      }
>>     } else {
>>        assert(chain->var->mode == vtn_variable_mode_ssbo);
>>        struct vtn_type *type;
>> --
>> 2.9.3
>>
>> _______________________________________________
>> mesa-dev mailing list
>> mesa-dev at lists.freedesktop.org
>> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
>>
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20160907/47bd8751/attachment-0001.html>