[Mesa-dev] [PATCH] nv50/ir: implement global atomics and handle it for nir

[Karol Herbst]

On Thu, Dec 5, 2019 at 11:57 AM Karol Herbst <kherbst at redhat.com> wrote:
>
> TGSI doesn't have any concept of global memory right now.
>
> Signed-off-by: Karol Herbst <kherbst at redhat.com>
> ---
>  .../nouveau/codegen/nv50_ir_from_nir.cpp      | 43 +++++++++++++++++--
>  .../nouveau/codegen/nv50_ir_lowering_nvc0.cpp |  2 +
>  2 files changed, 41 insertions(+), 4 deletions(-)
>
> diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_from_nir.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_from_nir.cpp
> index 08365988069..31f764d63d4 100644
> --- a/src/gallium/drivers/nouveau/codegen/nv50_ir_from_nir.cpp
> +++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_from_nir.cpp
> @@ -582,40 +582,47 @@ Converter::getSubOp(nir_intrinsic_op op)
>  {
>     switch (op) {
>     case nir_intrinsic_bindless_image_atomic_add:
> +   case nir_intrinsic_global_atomic_add:
>     case nir_intrinsic_image_atomic_add:
>     case nir_intrinsic_image_deref_atomic_add:
>     case nir_intrinsic_shared_atomic_add:
>     case nir_intrinsic_ssbo_atomic_add:
>        return  NV50_IR_SUBOP_ATOM_ADD;
>     case nir_intrinsic_bindless_image_atomic_and:
> +   case nir_intrinsic_global_atomic_and:
>     case nir_intrinsic_image_atomic_and:
>     case nir_intrinsic_image_deref_atomic_and:
>     case nir_intrinsic_shared_atomic_and:
>     case nir_intrinsic_ssbo_atomic_and:
>        return  NV50_IR_SUBOP_ATOM_AND;
>     case nir_intrinsic_bindless_image_atomic_comp_swap:
> +   case nir_intrinsic_global_atomic_comp_swap:
>     case nir_intrinsic_image_atomic_comp_swap:
>     case nir_intrinsic_image_deref_atomic_comp_swap:
>     case nir_intrinsic_shared_atomic_comp_swap:
>     case nir_intrinsic_ssbo_atomic_comp_swap:
>        return  NV50_IR_SUBOP_ATOM_CAS;
>     case nir_intrinsic_bindless_image_atomic_exchange:
> +   case nir_intrinsic_global_atomic_exchange:
>     case nir_intrinsic_image_atomic_exchange:
>     case nir_intrinsic_image_deref_atomic_exchange:
>     case nir_intrinsic_shared_atomic_exchange:
>     case nir_intrinsic_ssbo_atomic_exchange:
>        return  NV50_IR_SUBOP_ATOM_EXCH;
>     case nir_intrinsic_bindless_image_atomic_or:
> +   case nir_intrinsic_global_atomic_or:
>     case nir_intrinsic_image_atomic_or:
>     case nir_intrinsic_image_deref_atomic_or:
>     case nir_intrinsic_shared_atomic_or:
>     case nir_intrinsic_ssbo_atomic_or:
>        return  NV50_IR_SUBOP_ATOM_OR;
>     case nir_intrinsic_bindless_image_atomic_imax:
> -   case nir_intrinsic_image_atomic_imax:
> -   case nir_intrinsic_image_deref_atomic_imax:
>     case nir_intrinsic_bindless_image_atomic_umax:
> +   case nir_intrinsic_global_atomic_imax:
> +   case nir_intrinsic_global_atomic_umax:
> +   case nir_intrinsic_image_atomic_imax:
>     case nir_intrinsic_image_atomic_umax:
> +   case nir_intrinsic_image_deref_atomic_imax:
>     case nir_intrinsic_image_deref_atomic_umax:
>     case nir_intrinsic_shared_atomic_imax:
>     case nir_intrinsic_shared_atomic_umax:
> @@ -623,10 +630,12 @@ Converter::getSubOp(nir_intrinsic_op op)
>     case nir_intrinsic_ssbo_atomic_umax:
>        return  NV50_IR_SUBOP_ATOM_MAX;
>     case nir_intrinsic_bindless_image_atomic_imin:
> -   case nir_intrinsic_image_atomic_imin:
> -   case nir_intrinsic_image_deref_atomic_imin:
>     case nir_intrinsic_bindless_image_atomic_umin:
> +   case nir_intrinsic_global_atomic_imin:
> +   case nir_intrinsic_global_atomic_umin:
> +   case nir_intrinsic_image_atomic_imin:
>     case nir_intrinsic_image_atomic_umin:
> +   case nir_intrinsic_image_deref_atomic_imin:
>     case nir_intrinsic_image_deref_atomic_umin:
>     case nir_intrinsic_shared_atomic_imin:
>     case nir_intrinsic_shared_atomic_umin:
> @@ -634,6 +643,7 @@ Converter::getSubOp(nir_intrinsic_op op)
>     case nir_intrinsic_ssbo_atomic_umin:
>        return  NV50_IR_SUBOP_ATOM_MIN;
>     case nir_intrinsic_bindless_image_atomic_xor:
> +   case nir_intrinsic_global_atomic_xor:
>     case nir_intrinsic_image_atomic_xor:
>     case nir_intrinsic_image_deref_atomic_xor:
>     case nir_intrinsic_shared_atomic_xor:
> @@ -2379,6 +2389,30 @@ Converter::visit(nir_intrinsic_instr *insn)
>        info->io.globalAccess |= 0x2;
>        break;
>     }
> +   case nir_intrinsic_global_atomic_add:
> +   case nir_intrinsic_global_atomic_and:
> +   case nir_intrinsic_global_atomic_comp_swap:
> +   case nir_intrinsic_global_atomic_exchange:
> +   case nir_intrinsic_global_atomic_or:
> +   case nir_intrinsic_global_atomic_imax:
> +   case nir_intrinsic_global_atomic_imin:
> +   case nir_intrinsic_global_atomic_umax:
> +   case nir_intrinsic_global_atomic_umin:
> +   case nir_intrinsic_global_atomic_xor: {
> +      const DataType dType = getDType(insn);
> +      LValues &newDefs = convert(&insn->dest);
> +      Value *address;
> +      uint32_t offset = getIndirect(&insn->src[0], 0, address);
> +
> +      Symbol *sym = mkSymbol(FILE_MEMORY_GLOBAL, 0, dType, offset);
> +      Instruction *atom =
> +         mkOp2(OP_ATOM, dType, newDefs[0], sym, getSrc(&insn->src[1], 0));
> +      atom->setIndirect(0, 0, address);
> +      atom->subOp = getSubOp(op);
> +
> +      info->io.globalAccess |= 0x2;
> +      break;
> +   }
>     case nir_intrinsic_bindless_image_atomic_add:
>     case nir_intrinsic_bindless_image_atomic_and:
>     case nir_intrinsic_bindless_image_atomic_comp_swap:
> @@ -2860,6 +2894,7 @@ Converter::visit(nir_alu_instr *insn)
>     case nir_op_i2f32:
>     case nir_op_i2i32:
>     case nir_op_u2f32:
> +   case nir_op_u2u8:

ehh, this change actually slipped in and I wanted to remove it from
this patch...

>     case nir_op_u2u32:
>     case nir_op_f2f64:
>     case nir_op_f2i64:
> 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 a76d6c60cda..a60881000fe 100644
> --- a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
> +++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
> @@ -1645,6 +1645,8 @@ NVC0LoweringPass::handleATOM(Instruction *atom)
>        else if (targ->getChipset() < NVISA_GM107_CHIPSET)
>           handleSharedATOMNVE4(atom);
>        return true;
> +   case FILE_MEMORY_GLOBAL:
> +      return true;
>     default:
>        assert(atom->src(0).getFile() == FILE_MEMORY_BUFFER);
>        base = loadBufInfo64(ind, atom->getSrc(0)->reg.fileIndex * 16);
> --
> 2.23.0
>