Merge pull request #1748 from billhollings/fastmath-nocontraction
MSL: Honor DecorationNoContraction when compiling using fast-math.
diff --git a/reference/opt/shaders-msl/vert/float-math.invariant-float-math.vert b/reference/opt/shaders-msl/vert/float-math.invariant-float-math.vert
index 05e09e2..d8f44be 100644
--- a/reference/opt/shaders-msl/vert/float-math.invariant-float-math.vert
+++ b/reference/opt/shaders-msl/vert/float-math.invariant-float-math.vert
@@ -69,13 +69,13 @@
};
template<typename T>
-T spvFMul(T l, T r)
+[[clang::optnone]] T spvFMul(T l, T r)
{
return fma(l, r, T(0));
}
template<typename T, int Cols, int Rows>
-vec<T, Cols> spvFMulVectorMatrix(vec<T, Rows> v, matrix<T, Cols, Rows> m)
+[[clang::optnone]] vec<T, Cols> spvFMulVectorMatrix(vec<T, Rows> v, matrix<T, Cols, Rows> m)
{
vec<T, Cols> res = vec<T, Cols>(0);
for (uint i = Rows; i > 0; --i)
@@ -91,7 +91,7 @@
}
template<typename T, int Cols, int Rows>
-vec<T, Rows> spvFMulMatrixVector(matrix<T, Cols, Rows> m, vec<T, Cols> v)
+[[clang::optnone]] vec<T, Rows> spvFMulMatrixVector(matrix<T, Cols, Rows> m, vec<T, Cols> v)
{
vec<T, Rows> res = vec<T, Rows>(0);
for (uint i = Cols; i > 0; --i)
@@ -102,7 +102,7 @@
}
template<typename T, int LCols, int LRows, int RCols, int RRows>
-matrix<T, RCols, LRows> spvFMulMatrixMatrix(matrix<T, LCols, LRows> l, matrix<T, RCols, RRows> r)
+[[clang::optnone]] matrix<T, RCols, LRows> spvFMulMatrixMatrix(matrix<T, LCols, LRows> l, matrix<T, RCols, RRows> r)
{
matrix<T, RCols, LRows> res;
for (uint i = 0; i < RCols; i++)
diff --git a/reference/opt/shaders-msl/vert/no-contraction.vert b/reference/opt/shaders-msl/vert/no-contraction.vert
index a48731e..0b75dbc 100644
--- a/reference/opt/shaders-msl/vert/no-contraction.vert
+++ b/reference/opt/shaders-msl/vert/no-contraction.vert
@@ -18,13 +18,13 @@
};
template<typename T>
-T spvFMul(T l, T r)
+[[clang::optnone]] T spvFMul(T l, T r)
{
return fma(l, r, T(0));
}
template<typename T, int Cols, int Rows>
-vec<T, Cols> spvFMulVectorMatrix(vec<T, Rows> v, matrix<T, Cols, Rows> m)
+[[clang::optnone]] vec<T, Cols> spvFMulVectorMatrix(vec<T, Rows> v, matrix<T, Cols, Rows> m)
{
vec<T, Cols> res = vec<T, Cols>(0);
for (uint i = Rows; i > 0; --i)
@@ -40,7 +40,7 @@
}
template<typename T, int Cols, int Rows>
-vec<T, Rows> spvFMulMatrixVector(matrix<T, Cols, Rows> m, vec<T, Cols> v)
+[[clang::optnone]] vec<T, Rows> spvFMulMatrixVector(matrix<T, Cols, Rows> m, vec<T, Cols> v)
{
vec<T, Rows> res = vec<T, Rows>(0);
for (uint i = Cols; i > 0; --i)
@@ -51,7 +51,7 @@
}
template<typename T, int LCols, int LRows, int RCols, int RRows>
-matrix<T, RCols, LRows> spvFMulMatrixMatrix(matrix<T, LCols, LRows> l, matrix<T, RCols, RRows> r)
+[[clang::optnone]] matrix<T, RCols, LRows> spvFMulMatrixMatrix(matrix<T, LCols, LRows> l, matrix<T, RCols, RRows> r)
{
matrix<T, RCols, LRows> res;
for (uint i = 0; i < RCols; i++)
@@ -67,13 +67,13 @@
}
template<typename T>
-T spvFAdd(T l, T r)
+[[clang::optnone]] T spvFAdd(T l, T r)
{
return fma(T(1), l, r);
}
template<typename T>
-T spvFSub(T l, T r)
+[[clang::optnone]] T spvFSub(T l, T r)
{
return fma(T(-1), r, l);
}
diff --git a/reference/shaders-msl/vert/float-math.invariant-float-math.vert b/reference/shaders-msl/vert/float-math.invariant-float-math.vert
index d603884..06844ca 100644
--- a/reference/shaders-msl/vert/float-math.invariant-float-math.vert
+++ b/reference/shaders-msl/vert/float-math.invariant-float-math.vert
@@ -69,13 +69,13 @@
};
template<typename T>
-T spvFMul(T l, T r)
+[[clang::optnone]] T spvFMul(T l, T r)
{
return fma(l, r, T(0));
}
template<typename T, int Cols, int Rows>
-vec<T, Cols> spvFMulVectorMatrix(vec<T, Rows> v, matrix<T, Cols, Rows> m)
+[[clang::optnone]] vec<T, Cols> spvFMulVectorMatrix(vec<T, Rows> v, matrix<T, Cols, Rows> m)
{
vec<T, Cols> res = vec<T, Cols>(0);
for (uint i = Rows; i > 0; --i)
@@ -91,7 +91,7 @@
}
template<typename T, int Cols, int Rows>
-vec<T, Rows> spvFMulMatrixVector(matrix<T, Cols, Rows> m, vec<T, Cols> v)
+[[clang::optnone]] vec<T, Rows> spvFMulMatrixVector(matrix<T, Cols, Rows> m, vec<T, Cols> v)
{
vec<T, Rows> res = vec<T, Rows>(0);
for (uint i = Cols; i > 0; --i)
@@ -102,7 +102,7 @@
}
template<typename T, int LCols, int LRows, int RCols, int RRows>
-matrix<T, RCols, LRows> spvFMulMatrixMatrix(matrix<T, LCols, LRows> l, matrix<T, RCols, RRows> r)
+[[clang::optnone]] matrix<T, RCols, LRows> spvFMulMatrixMatrix(matrix<T, LCols, LRows> l, matrix<T, RCols, RRows> r)
{
matrix<T, RCols, LRows> res;
for (uint i = 0; i < RCols; i++)
diff --git a/reference/shaders-msl/vert/no-contraction.vert b/reference/shaders-msl/vert/no-contraction.vert
index 907d901..653dc26 100644
--- a/reference/shaders-msl/vert/no-contraction.vert
+++ b/reference/shaders-msl/vert/no-contraction.vert
@@ -18,13 +18,13 @@
};
template<typename T>
-T spvFMul(T l, T r)
+[[clang::optnone]] T spvFMul(T l, T r)
{
return fma(l, r, T(0));
}
template<typename T, int Cols, int Rows>
-vec<T, Cols> spvFMulVectorMatrix(vec<T, Rows> v, matrix<T, Cols, Rows> m)
+[[clang::optnone]] vec<T, Cols> spvFMulVectorMatrix(vec<T, Rows> v, matrix<T, Cols, Rows> m)
{
vec<T, Cols> res = vec<T, Cols>(0);
for (uint i = Rows; i > 0; --i)
@@ -40,7 +40,7 @@
}
template<typename T, int Cols, int Rows>
-vec<T, Rows> spvFMulMatrixVector(matrix<T, Cols, Rows> m, vec<T, Cols> v)
+[[clang::optnone]] vec<T, Rows> spvFMulMatrixVector(matrix<T, Cols, Rows> m, vec<T, Cols> v)
{
vec<T, Rows> res = vec<T, Rows>(0);
for (uint i = Cols; i > 0; --i)
@@ -51,7 +51,7 @@
}
template<typename T, int LCols, int LRows, int RCols, int RRows>
-matrix<T, RCols, LRows> spvFMulMatrixMatrix(matrix<T, LCols, LRows> l, matrix<T, RCols, RRows> r)
+[[clang::optnone]] matrix<T, RCols, LRows> spvFMulMatrixMatrix(matrix<T, LCols, LRows> l, matrix<T, RCols, RRows> r)
{
matrix<T, RCols, LRows> res;
for (uint i = 0; i < RCols; i++)
@@ -67,13 +67,13 @@
}
template<typename T>
-T spvFAdd(T l, T r)
+[[clang::optnone]] T spvFAdd(T l, T r)
{
return fma(T(1), l, r);
}
template<typename T>
-T spvFSub(T l, T r)
+[[clang::optnone]] T spvFSub(T l, T r)
{
return fma(T(-1), r, l);
}
diff --git a/spirv_msl.cpp b/spirv_msl.cpp
index 9e797f6..2b1e062 100644
--- a/spirv_msl.cpp
+++ b/spirv_msl.cpp
@@ -4928,7 +4928,7 @@
// "fadd" intrinsic support
case SPVFuncImplFAdd:
statement("template<typename T>");
- statement("T spvFAdd(T l, T r)");
+ statement("[[clang::optnone]] T spvFAdd(T l, T r)");
begin_scope();
statement("return fma(T(1), l, r);");
end_scope();
@@ -4938,7 +4938,7 @@
// "fsub" intrinsic support
case SPVFuncImplFSub:
statement("template<typename T>");
- statement("T spvFSub(T l, T r)");
+ statement("[[clang::optnone]] T spvFSub(T l, T r)");
begin_scope();
statement("return fma(T(-1), r, l);");
end_scope();
@@ -4948,14 +4948,14 @@
// "fmul' intrinsic support
case SPVFuncImplFMul:
statement("template<typename T>");
- statement("T spvFMul(T l, T r)");
+ statement("[[clang::optnone]] T spvFMul(T l, T r)");
begin_scope();
statement("return fma(l, r, T(0));");
end_scope();
statement("");
statement("template<typename T, int Cols, int Rows>");
- statement("vec<T, Cols> spvFMulVectorMatrix(vec<T, Rows> v, matrix<T, Cols, Rows> m)");
+ statement("[[clang::optnone]] vec<T, Cols> spvFMulVectorMatrix(vec<T, Rows> v, matrix<T, Cols, Rows> m)");
begin_scope();
statement("vec<T, Cols> res = vec<T, Cols>(0);");
statement("for (uint i = Rows; i > 0; --i)");
@@ -4972,7 +4972,7 @@
statement("");
statement("template<typename T, int Cols, int Rows>");
- statement("vec<T, Rows> spvFMulMatrixVector(matrix<T, Cols, Rows> m, vec<T, Cols> v)");
+ statement("[[clang::optnone]] vec<T, Rows> spvFMulMatrixVector(matrix<T, Cols, Rows> m, vec<T, Cols> v)");
begin_scope();
statement("vec<T, Rows> res = vec<T, Rows>(0);");
statement("for (uint i = Cols; i > 0; --i)");
@@ -4984,8 +4984,7 @@
statement("");
statement("template<typename T, int LCols, int LRows, int RCols, int RRows>");
- statement(
- "matrix<T, RCols, LRows> spvFMulMatrixMatrix(matrix<T, LCols, LRows> l, matrix<T, RCols, RRows> r)");
+ statement("[[clang::optnone]] matrix<T, RCols, LRows> spvFMulMatrixMatrix(matrix<T, LCols, LRows> l, matrix<T, RCols, RRows> r)");
begin_scope();
statement("matrix<T, RCols, LRows> res;");
statement("for (uint i = 0; i < RCols; i++)");
