Honor NoContraction qualifier.

We'll need to force a temporary and mark it as precise.
MSL is a little weird here, but we can piggyback on top of the invariant
float math option here to force fma() operations everywhere.

3 files changed, 145 insertions, 0 deletions

diff --git a/reference/opt/shaders-hlsl/vert/no-contraction.vert b/reference/opt/shaders-hlsl/vert/no-contraction.vert
new file mode 100644
index 00000000..10763fbe
--- /dev/null
+++ b/reference/opt/shaders-hlsl/vert/no-contraction.vert
@@ -0,0 +1,39 @@
+static float4 gl_Position;
+static float4 vA;
+static float4 vB;
+static float4 vC;
+
+struct SPIRV_Cross_Input
+{
+    float4 vA : TEXCOORD0;
+    float4 vB : TEXCOORD1;
+    float4 vC : TEXCOORD2;
+};
+
+struct SPIRV_Cross_Output
+{
+    float4 gl_Position : SV_Position;
+};
+
+void vert_main()
+{
+    precise float4 _15 = vA * vB;
+    precise float4 _19 = vA + vB;
+    precise float4 _23 = vA - vB;
+    precise float4 _30 = _15 + vC;
+    precise float4 _34 = _15 + _19;
+    precise float4 _36 = _34 + _23;
+    precise float4 _38 = _36 + _30;
+    gl_Position = _38;
+}
+
+SPIRV_Cross_Output main(SPIRV_Cross_Input stage_input)
+{
+    vA = stage_input.vA;
+    vB = stage_input.vB;
+    vC = stage_input.vC;
+    vert_main();
+    SPIRV_Cross_Output stage_output;
+    stage_output.gl_Position = gl_Position;
+    return stage_output;
+}
diff --git a/reference/opt/shaders-msl/vert/no-contraction.vert b/reference/opt/shaders-msl/vert/no-contraction.vert
new file mode 100644
index 00000000..a48731ed
--- /dev/null
+++ b/reference/opt/shaders-msl/vert/no-contraction.vert
@@ -0,0 +1,88 @@
+#pragma clang diagnostic ignored "-Wmissing-prototypes"
+
+#include <metal_stdlib>
+#include <simd/simd.h>
+
+using namespace metal;
+
+struct main0_out
+{
+    float4 gl_Position [[position]];
+};
+
+struct main0_in
+{
+    float4 vA [[attribute(0)]];
+    float4 vB [[attribute(1)]];
+    float4 vC [[attribute(2)]];
+};
+
+template<typename T>
+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)
+{
+    vec<T, Cols> res = vec<T, Cols>(0);
+    for (uint i = Rows; i > 0; --i)
+    {
+        vec<T, Cols> tmp(0);
+        for (uint j = 0; j < Cols; ++j)
+        {
+            tmp[j] = m[j][i - 1];
+        }
+        res = fma(tmp, vec<T, Cols>(v[i - 1]), res);
+    }
+    return res;
+}
+
+template<typename T, int Cols, int Rows>
+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)
+    {
+        res = fma(m[i - 1], vec<T, Rows>(v[i - 1]), res);
+    }
+    return res;
+}
+
+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)
+{
+    matrix<T, RCols, LRows> res;
+    for (uint i = 0; i < RCols; i++)
+    {
+        vec<T, RCols> tmp(0);
+        for (uint j = 0; j < LCols; j++)
+        {
+            tmp = fma(vec<T, RCols>(r[i][j]), l[j], tmp);
+        }
+        res[i] = tmp;
+    }
+    return res;
+}
+
+template<typename T>
+T spvFAdd(T l, T r)
+{
+    return fma(T(1), l, r);
+}
+
+template<typename T>
+T spvFSub(T l, T r)
+{
+    return fma(T(-1), r, l);
+}
+
+vertex main0_out main0(main0_in in [[stage_in]])
+{
+    main0_out out = {};
+    float4 _15 = spvFMul(in.vA, in.vB);
+    out.gl_Position = spvFAdd(spvFAdd(spvFAdd(_15, spvFAdd(in.vA, in.vB)), spvFSub(in.vA, in.vB)), spvFAdd(_15, in.vC));
+    return out;
+}
+
diff --git a/reference/opt/shaders/vert/no-contraction.vert b/reference/opt/shaders/vert/no-contraction.vert
new file mode 100644
index 00000000..9f9969cd
--- /dev/null
+++ b/reference/opt/shaders/vert/no-contraction.vert
@@ -0,0 +1,18 @@
+#version 450
+
+layout(location = 0) in vec4 vA;
+layout(location = 1) in vec4 vB;
+layout(location = 2) in vec4 vC;
+
+void main()
+{
+    precise vec4 _15 = vA * vB;
+    precise vec4 _19 = vA + vB;
+    precise vec4 _23 = vA - vB;
+    precise vec4 _30 = _15 + vC;
+    precise vec4 _34 = _15 + _19;
+    precise vec4 _36 = _34 + _23;
+    precise vec4 _38 = _36 + _30;
+    gl_Position = _38;
+}
+