[Mesa-stable] [PATCH] ac/nir: mark some texture intrinsics as convergent
Marek Olšák
maraeo at gmail.com
Fri May 31 19:07:49 UTC 2019
I see. Rb for the whole patch then.
Marek
On Fri, May 31, 2019, 2:24 PM Rhys Perry <pendingchaos02 at gmail.com> wrote:
> The first and last hunks are needed to pass on the shader_info to the
> middle hunk, which needs it so that it can test if the compute shader
> has a derivative group.
>
> On Fri, 31 May 2019 at 18:38, Marek Olšák <maraeo at gmail.com> wrote:
> >
> > The first and last hunks look like they shouldn't be there. Other than
> that:
> >
> > Reviewed-by: Marek Olšák <marek.olsak at amd.com>
> >
> > Marek
> >
> > On Fri, May 31, 2019 at 11:53 AM Rhys Perry <pendingchaos02 at gmail.com>
> wrote:
> >>
> >> Otherwise LLVM can sink them and their texture coordinate calculations
> >> into divergent branches.
> >>
> >> v2: simplify the conditions on which the intrinsic is marked as
> convergent
> >> v3: only mark as convergent in FS and CS with derivative groups
> >>
> >> Cc: <mesa-stable at lists.freedesktop.org>
> >> Signed-off-by: Rhys Perry <pendingchaos02 at gmail.com>
> >> ---
> >> src/amd/common/ac_nir_to_llvm.c | 18 ++++++++++++++++++
> >> 1 file changed, 18 insertions(+)
> >>
> >> diff --git a/src/amd/common/ac_nir_to_llvm.c
> b/src/amd/common/ac_nir_to_llvm.c
> >> index 265e3b636c4..9e9fade7227 100644
> >> --- a/src/amd/common/ac_nir_to_llvm.c
> >> +++ b/src/amd/common/ac_nir_to_llvm.c
> >> @@ -38,6 +38,7 @@ struct ac_nir_context {
> >> struct ac_shader_abi *abi;
> >>
> >> gl_shader_stage stage;
> >> + shader_info *info;
> >>
> >> LLVMValueRef *ssa_defs;
> >>
> >> @@ -1394,6 +1395,22 @@ static LLVMValueRef build_tex_intrinsic(struct
> ac_nir_context *ctx,
> >> }
> >>
> >> args->attributes = AC_FUNC_ATTR_READNONE;
> >> + bool cs_derivs = ctx->stage == MESA_SHADER_COMPUTE &&
> >> + ctx->info->cs.derivative_group !=
> DERIVATIVE_GROUP_NONE;
> >> + if (ctx->stage == MESA_SHADER_FRAGMENT || cs_derivs) {
> >> + /* Prevent texture instructions with implicit
> derivatives from being
> >> + * sinked into branches. */
> >> + switch (instr->op) {
> >> + case nir_texop_tex:
> >> + case nir_texop_txb:
> >> + case nir_texop_lod:
> >> + args->attributes |= AC_FUNC_ATTR_CONVERGENT;
> >> + break;
> >> + default:
> >> + break;
> >> + }
> >> + }
> >> +
> >> return ac_build_image_opcode(&ctx->ac, args);
> >> }
> >>
> >> @@ -4350,6 +4367,7 @@ void ac_nir_translate(struct ac_llvm_context *ac,
> struct ac_shader_abi *abi,
> >> ctx.abi = abi;
> >>
> >> ctx.stage = nir->info.stage;
> >> + ctx.info = &nir->info;
> >>
> >> ctx.main_function = LLVMGetBasicBlockParent(LLVMGetInsertBlock(
> ctx.ac.builder));
> >>
> >> --
> >> 2.21.0
> >>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/mesa-stable/attachments/20190531/df83d3cd/attachment-0001.html>
More information about the mesa-stable
mailing list