Pierre Moreau <pierre.mor...@free.fr> writes: > On 2018-01-23 — 14:07, Jan Vesely wrote: >> On Tue, 2018-01-23 at 01:33 +0100, Pierre Moreau wrote: >> > Signed-off-by: Pierre Moreau <pierre.mor...@free.fr> >> > --- >> > src/gallium/include/pipe/p_defines.h | 1 + >> > 1 file changed, 1 insertion(+) >> > >> > diff --git a/src/gallium/include/pipe/p_defines.h >> > b/src/gallium/include/pipe/p_defines.h >> > index b34e7a8570..082d4c4d87 100644 >> > --- a/src/gallium/include/pipe/p_defines.h >> > +++ b/src/gallium/include/pipe/p_defines.h >> > @@ -896,6 +896,7 @@ enum pipe_shader_ir >> > PIPE_SHADER_IR_LLVM, >> > PIPE_SHADER_IR_NATIVE, >> > PIPE_SHADER_IR_NIR, >> > + PIPE_SHADER_IR_SPIRV >> >> Why is this needed/useful? presumably the pipe driver will convert >> SPIRV to NIR or LLVM anyway, why not convert it in clover and pass one >> of the already existing IRs ? >> >> Jan > > It is true it’s not really needed for this series, it’s mainly for my SPIR-V > frontend for Nouveau. I could drop it from this series for now, especially if > SPIR-V get directly translated to another IR and never stays as such, or going > the opposite and always keeping SPIR-V until the very last moment. >
I don't have any objection against plumbing through SPIRV directly if that's the path you're planning to pursue. But wouldn't it make sense to have working back-end code able to consume it as input before adding a new IR to the pipe driver interface? Otherwise this is basically dead code. > Pierre > >> >> > }; >> > >> > /**
signature.asc
Description: PGP signature
_______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev
