[Mesa-dev] [PATCH] swr: simd16 vs work

[Roland Scheidegger]

FWIW I think it would make a lot more sense if you'd just extend the
gallivm code to be able to build proper simd16 shaders.
But just my 2 cents...

Roland

Am 19.04.2017 um 15:41 schrieb Tim Rowley:
> Build VS with alternating output for the current simd16 fe double-pump
> of a simd8 shader.
> ---
>  src/gallium/drivers/swr/swr_shader.cpp | 30 +++++++++++++++++++++++++-----
>  1 file changed, 25 insertions(+), 5 deletions(-)
> 
> diff --git a/src/gallium/drivers/swr/swr_shader.cpp b/src/gallium/drivers/swr/swr_shader.cpp
> index d8f5512..ec38584 100644
> --- a/src/gallium/drivers/swr/swr_shader.cpp
> +++ b/src/gallium/drivers/swr/swr_shader.cpp
> @@ -226,6 +226,9 @@ struct BuilderSWR : public Builder {
>        gallivm_free_ir(gallivm);
>     }
>  
> +   void WriteVS(Value *pVal, Value *pVsContext, Value *pVtxOutput,
> +                unsigned slot, unsigned channel);
> +
>     struct gallivm_state *gallivm;
>     PFN_VERTEX_FUNC CompileVS(struct swr_context *ctx, swr_jit_vs_key &key);
>     PFN_PIXEL_KERNEL CompileFS(struct swr_context *ctx, swr_jit_fs_key &key);
> @@ -657,6 +660,23 @@ swr_compile_gs(struct swr_context *ctx, swr_jit_gs_key &key)
>     return func;
>  }
>  
> +void
> +BuilderSWR::WriteVS(Value *pVal, Value *pVsContext, Value *pVtxOutput, unsigned slot, unsigned channel)
> +{
> +#if USE_SIMD16_FRONTEND
> +   // interleave the simdvertex components into the dest simd16vertex
> +   //   slot16offset = slot8offset * 2
> +   //   comp16offset = comp8offset * 2 + alternateOffset
> +
> +   Value *offset = LOAD(pVsContext, { 0, SWR_VS_CONTEXT_AlternateOffset });
> +   Value *pOut = GEP(pVtxOutput, { C(0), C(0), C(slot * 2), offset } );
> +   STORE(pVal, pOut, {channel * 2});
> +#else
> +   Value *pOut = GEP(pVtxOutput, {0, 0, slot});
> +   STORE(pVal, pOut, {0, channel});
> +#endif
> +}
> +
>  PFN_VERTEX_FUNC
>  BuilderSWR::CompileVS(struct swr_context *ctx, swr_jit_vs_key &key)
>  {
> @@ -752,7 +772,7 @@ BuilderSWR::CompileVS(struct swr_context *ctx, swr_jit_vs_key &key)
>           uint32_t outSlot = attrib;
>           if (swr_vs->info.base.output_semantic_name[attrib] == TGSI_SEMANTIC_PSIZE)
>              outSlot = VERTEX_POINT_SIZE_SLOT;
> -         STORE(val, vtxOutput, {0, 0, outSlot, channel});
> +         WriteVS(val, pVsCtx, vtxOutput, outSlot, channel);
>        }
>     }
>  
> @@ -786,10 +806,10 @@ BuilderSWR::CompileVS(struct swr_context *ctx, swr_jit_vs_key &key)
>                                               &swr_vs->info.base);
>              if (val < 4) {
>                 LLVMValueRef dist = LLVMBuildLoad(gallivm->builder, outputs[cv][val], "");
> -               STORE(unwrap(dist), vtxOutput, {0, 0, VERTEX_CLIPCULL_DIST_LO_SLOT, val});
> +               WriteVS(unwrap(dist), pVsCtx, vtxOutput, VERTEX_CLIPCULL_DIST_LO_SLOT, val);
>              } else {
>                 LLVMValueRef dist = LLVMBuildLoad(gallivm->builder, outputs[cv][val - 4], "");
> -               STORE(unwrap(dist), vtxOutput, {0, 0, VERTEX_CLIPCULL_DIST_HI_SLOT, val - 4});
> +               WriteVS(unwrap(dist), pVsCtx, vtxOutput, VERTEX_CLIPCULL_DIST_HI_SLOT, val - 4);
>              }
>              continue;
>           }
> @@ -807,9 +827,9 @@ BuilderSWR::CompileVS(struct swr_context *ctx, swr_jit_vs_key &key)
>                                        FMUL(unwrap(cw), VBROADCAST(pw)))));
>  
>           if (val < 4)
> -            STORE(dist, vtxOutput, {0, 0, VERTEX_CLIPCULL_DIST_LO_SLOT, val});
> +            WriteVS(dist, pVsCtx, vtxOutput, VERTEX_CLIPCULL_DIST_LO_SLOT, val);
>           else
> -            STORE(dist, vtxOutput, {0, 0, VERTEX_CLIPCULL_DIST_HI_SLOT, val - 4});
> +            WriteVS(dist, pVsCtx, vtxOutput, VERTEX_CLIPCULL_DIST_HI_SLOT, val - 4);
>        }
>     }
>  
>