lux/yuzu
1
0
Fork 0
mirror of https://git.h3cjp.net/H3cJP/yuzu.git synced 2024-12-18 16:56:19 +00:00
Code Issues Packages Projects Releases Wiki Activity

glsl: Rework Shuffle emit instructions to align with SPIR-V

Browse source
This commit is contained in:
ameerj 2021-05-28 21:58:26 -04:00
parent 8ba814efb2
commit b98de76ea8
1 changed files with 40 additions and 19 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

59
src/shader_recompiler/backend/glsl/emit_glsl_warp.cpp
View file

@ -14,51 +14,72 @@ void SetInBoundsFlag(EmitContext& ctx, IR::Inst& inst) {
if (!in_bounds) {
return;
}
ctx.AddU1("{}=shfl_in_bounds;", *in_bounds);
in_bounds->Invalidate();
}
std::string ComputeMinThreadId(std::string_view thread_id, std::string_view segmentation_mask) {
return fmt::format("({}&{})", thread_id, segmentation_mask);
}
std::string ComputeMaxThreadId(std::string_view min_thread_id, std::string_view clamp,
std::string_view not_seg_mask) {
return fmt::format("({})|({}&{})", min_thread_id, clamp, not_seg_mask);
}
std::string GetMaxThreadId(std::string_view thread_id, std::string_view clamp,
std::string_view segmentation_mask) {
const auto not_seg_mask{fmt::format("(~{})", segmentation_mask)};
const auto min_thread_id{ComputeMinThreadId(thread_id, segmentation_mask)};
return ComputeMaxThreadId(min_thread_id, clamp, not_seg_mask);
}
} // namespace
void EmitShuffleIndex(EmitContext& ctx, IR::Inst& inst, std::string_view value,
std::string_view index, std::string_view clamp,
std::string_view segmentation_mask) {
ctx.Add("shfl_in_bounds=int(gl_SubGroupInvocationARB-{})>=int((gl_SubGroupInvocationARB&{})|({}"
"&~{}));",
index, segmentation_mask, clamp, segmentation_mask);
const auto not_seg_mask{fmt::format("(~{})", segmentation_mask)};
const auto thread_id{"gl_SubGroupInvocationARB"};
const auto min_thread_id{ComputeMinThreadId(thread_id, segmentation_mask)};
const auto max_thread_id{ComputeMaxThreadId(min_thread_id, clamp, not_seg_mask)};
const auto lhs{fmt::format("({}&{})", index, not_seg_mask)};
const auto src_thread_id{fmt::format("({})|({})", lhs, min_thread_id)};
ctx.Add("shfl_in_bounds=int({})<=int({});", src_thread_id, max_thread_id);
SetInBoundsFlag(ctx, inst);
ctx.AddU32("{}=shfl_in_bounds?{}:gl_SubGroupInvocationARB-{};", inst, value, index);
ctx.AddU32("{}=shfl_in_bounds?{}:{};", inst, value, src_thread_id);
}
void EmitShuffleUp(EmitContext& ctx, IR::Inst& inst, std::string_view value, std::string_view index,
std::string_view clamp, std::string_view segmentation_mask) {
ctx.Add("shfl_in_bounds=int(gl_SubGroupInvocationARB-{})>=int((gl_SubGroupInvocationARB&{})|({}"
"&~{}));",
index, segmentation_mask, clamp, segmentation_mask);
const auto thread_id{"gl_SubGroupInvocationARB"};
const auto max_thread_id{GetMaxThreadId(thread_id, clamp, segmentation_mask)};
const auto src_thread_id{fmt::format("({}-{})", thread_id, index)};
ctx.Add("shfl_in_bounds=int({})>=int({});", src_thread_id, max_thread_id);
SetInBoundsFlag(ctx, inst);
ctx.AddU32("{}=shfl_in_bounds?readInvocationARB({},gl_SubGroupInvocationARB-{}):"
"{};",
inst, value, index, value);
ctx.AddU32("{}=shfl_in_bounds?{}:{};", inst, value, src_thread_id);
}
void EmitShuffleDown(EmitContext& ctx, IR::Inst& inst, std::string_view value,
std::string_view index, std::string_view clamp,
std::string_view segmentation_mask) {
ctx.Add("shfl_in_bounds=int(gl_SubGroupInvocationARB-{})>=int((gl_SubGroupInvocationARB&{})|({}"
"&~{}));",
index, segmentation_mask, clamp, segmentation_mask);
const auto thread_id{"gl_SubGroupInvocationARB"};
const auto max_thread_id{GetMaxThreadId(thread_id, clamp, segmentation_mask)};
const auto src_thread_id{fmt::format("({}+{})", thread_id, index)};
ctx.Add("shfl_in_bounds=int({})<=int({});", src_thread_id, max_thread_id);
SetInBoundsFlag(ctx, inst);
ctx.AddU32("{}=shfl_in_bounds?{}:gl_SubGroupInvocationARB-{};", inst, value, index);
ctx.AddU32("{}=shfl_in_bounds?{}:{};", inst, value, src_thread_id);
}
void EmitShuffleButterfly(EmitContext& ctx, IR::Inst& inst, std::string_view value,
std::string_view index, std::string_view clamp,
std::string_view segmentation_mask) {
ctx.Add("shfl_in_bounds=int(gl_SubGroupInvocationARB-{})>=int((gl_SubGroupInvocationARB&{})|({}"
"&~{}));",
index, segmentation_mask, clamp, segmentation_mask);
const auto thread_id{"gl_SubGroupInvocationARB"};
const auto max_thread_id{GetMaxThreadId(thread_id, clamp, segmentation_mask)};
const auto src_thread_id{fmt::format("({}^{})", thread_id, index)};
ctx.Add("shfl_in_bounds=int({})<=int({});", src_thread_id, max_thread_id);
SetInBoundsFlag(ctx, inst);
ctx.AddU32("{}=shfl_in_bounds?{}:gl_SubGroupInvocationARB-{};", inst, value, index);
ctx.AddU32("{}=shfl_in_bounds?{}:{};", inst, value, src_thread_id);
}
void EmitFSwizzleAdd([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,