Skip to content

S
swiftshader
Commit 7ecf4735 by Ben Clayton
Options

SpirvShader: Refactor AllTrueMask() / AnyTrueMask()

Replace these with the more generic AndAll() and OrAll(). These can be used to clean up some existing code. Bug: b/133510501 Tests: dEQP-VK.subgroups.* Change-Id: I49f84f6c7088c1c810a0dfda5358a2cc4778f79d Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/35088 Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Reviewed-by: 's avatarChris Forbes <chrisforbes@google.com> Tested-by: 's avatarBen Clayton <bclayton@google.com>
parent ca2c5ef7
Showing with 15 additions and 17 deletions
src/Pipeline/SpirvShader.cpp
...@@ -45,18 +45,20 @@ namespace ...@@ -45,18 +45,20 @@ namespace
return rr::SignMask(~ints) != 0; return rr::SignMask(~ints) != 0;
} }
rr::RValue<sw::SIMD::UInt> AllTrueMask(rr::RValue<sw::SIMD::UInt> const &mask) template <typename T>
rr::RValue<T> AndAll(rr::RValue<T> const &mask)
{ {
sw::SIMD::UInt v1 = mask; // [x] [y] [z] [w] T v1 = mask; // [x] [y] [z] [w]
sw::SIMD::UInt v2 = v1.xzxz & v1.ywyw; // [xy] [zw] [xy] [zw] T v2 = v1.xzxz & v1.ywyw; // [xy] [zw] [xy] [zw]
return v2.xxxx & v2.yyyy; // [xyzw] [xyzw] [xyzw] [xyzw] return v2.xxxx & v2.yyyy; // [xyzw] [xyzw] [xyzw] [xyzw]
} }
rr::RValue<sw::SIMD::UInt> AnyTrueMask(rr::RValue<sw::SIMD::UInt> const &mask) template <typename T>
rr::RValue<T> OrAll(rr::RValue<T> const &mask)
{ {
sw::SIMD::UInt v1 = mask; // [x] [y] [z] [w] T v1 = mask; // [x] [y] [z] [w]
sw::SIMD::UInt v2 = v1.xzxz | v1.ywyw; // [xy] [zw] [xy] [zw] T v2 = v1.xzxz | v1.ywyw; // [xy] [zw] [xy] [zw]
return v2.xxxx | v2.yyyy; // [xyzw] [xyzw] [xyzw] [xyzw] return v2.xxxx | v2.yyyy; // [xyzw] [xyzw] [xyzw] [xyzw]
} }
rr::RValue<sw::SIMD::Float> Sign(rr::RValue<sw::SIMD::Float> const &val) rr::RValue<sw::SIMD::Float> Sign(rr::RValue<sw::SIMD::Float> const &val)
...@@ -6007,14 +6009,14 @@ namespace sw ...@@ -6007,14 +6009,14 @@ namespace sw
case spv::OpGroupNonUniformAll: case spv::OpGroupNonUniformAll:
{ {
GenericValue predicate(this, state, insn.word(4)); GenericValue predicate(this, state, insn.word(4));
dst.move(0, AllTrueMask(predicate.UInt(0) | ~As<SIMD::UInt>(state->activeLaneMask()))); dst.move(0, AndAll(predicate.UInt(0) | ~As<SIMD::UInt>(state->activeLaneMask())));
break; break;
} }
case spv::OpGroupNonUniformAny: case spv::OpGroupNonUniformAny:
{ {
GenericValue predicate(this, state, insn.word(4)); GenericValue predicate(this, state, insn.word(4));
dst.move(0, AnyTrueMask(predicate.UInt(0) & As<SIMD::UInt>(state->activeLaneMask()))); dst.move(0, OrAll(predicate.UInt(0) & As<SIMD::UInt>(state->activeLaneMask())));
break; break;
} }
...@@ -6032,7 +6034,7 @@ namespace sw ...@@ -6032,7 +6034,7 @@ namespace sw
{ {
filled |= filled.yzwx & inactive; // Populate inactive 'holes' with a live value filled |= filled.yzwx & inactive; // Populate inactive 'holes' with a live value
} }
res &= AllTrueMask(CmpEQ(filled.xyzw, filled.yzwx)); res &= AndAll(CmpEQ(filled.xyzw, filled.yzwx));
} }
dst.move(0, res); dst.move(0, res);
break; break;
...@@ -6046,9 +6048,7 @@ namespace sw ...@@ -6046,9 +6048,7 @@ namespace sw
auto mask = CmpEQ(id, SIMD::Int(0, 1, 2, 3)); auto mask = CmpEQ(id, SIMD::Int(0, 1, 2, 3));
for (auto i = 0u; i < type.sizeInComponents; i++) for (auto i = 0u; i < type.sizeInComponents; i++)
{ {
auto oneVal = SIMD::Int(value.Int(i) & mask); dst.move(i, OrAll(value.Int(i) & mask));
auto replVal = SIMD::Int(oneVal.xxzz | oneVal.yyww);
dst.move(i, replVal.xxyy | replVal.zzww);
} }
break; break;
} }
...@@ -6066,9 +6066,7 @@ namespace sw ...@@ -6066,9 +6066,7 @@ namespace sw
auto elect = active & ~(v0111 & (active.xxyz | active.xxxy | active.xxxx)); auto elect = active & ~(v0111 & (active.xxyz | active.xxxy | active.xxxx));
for (auto i = 0u; i < type.sizeInComponents; i++) for (auto i = 0u; i < type.sizeInComponents; i++)
{ {
auto oneVal = SIMD::Int(value.Int(i) & elect); dst.move(i, OrAll(value.Int(i) & elect));
auto replVal = SIMD::Int(oneVal.xxzz | oneVal.yyww);
dst.move(i, replVal.xxyy | replVal.zzww);
} }
break; break;
} }
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment