Support Metal math functions (shader-slang#4118)

* Support Metal math functions Closes shader-slang#4024 Note that Metal document says Metal doesn't support "double" type; "Metal does not support the double, long long, unsigned long long, and long double data types." According to Metal document, math functions are not defined for integer types. That leaves only two types to test: half and float. As a code clean up, __floatCast is replaced with __realCast. But I had to add a new signature that can convert from integer to float. Some of GLSL functions are moved to hlsl.meta.slang. For those functions, there isn't builtin functions for HLSL but there are for GLSL and Metal. "nextafter(T,T)" is currently not working because it requires Metal version 3.1 and we invoke metal compiler with a profile version lower than 3.1. * Changes based on review comments.
jkwak-work · May 7, 2024 · 997f040 · 997f040
1 parent 1b3a428
commit 997f040
Show file tree

Hide file tree

Showing 5 changed files with 1,638 additions and 330 deletions.
diff --git a/source/slang/core.meta.slang b/source/slang/core.meta.slang
@@ -421,6 +421,8 @@ __generic<T, let N : int> __intrinsic_op(select) vector<T,N> select(vector<bool,
 // Allow real-number types to be cast into each other
 __intrinsic_op($(kIROp_FloatCast))
     T __realCast<T : __BuiltinRealType, U : __BuiltinRealType>(U val);
+__intrinsic_op($(kIROp_CastIntToFloat))
+    T __realCast<T : __BuiltinRealType, U : __BuiltinIntegerType>(U val);
 __intrinsic_op($(kIROp_IntCast))
     T __intCast<T : __BuiltinType, U : __BuiltinType>(U val);
 ${{{{

diff --git a/source/slang/glsl.meta.slang b/source/slang/glsl.meta.slang
@@ -321,114 +321,6 @@ public vector<T,N> atan(vector<T,N> y, vector<T,N> x)
     return atan2(y, x);
 }
 
-__generic<T : __BuiltinFloatingPointType>
-[__readNone]
-[ForceInline]
-[require(cpp_cuda_glsl_hlsl_spirv, GLSL_130)]
-public T asinh(T x)
-{
-    __target_switch
-    {
-    case cpp: __intrinsic_asm "$P_asinh($0)";
-    case cuda: __intrinsic_asm "$P_asinh($0)";
-    case glsl: __intrinsic_asm "asinh";
-    case spirv: return spirv_asm {
-        OpExtInst $$T result glsl450 Asinh $x
-    };
-    default:
-        return log(x + sqrt(x * x + T(1)));
-    }
-}
-
-__generic<T : __BuiltinFloatingPointType, let N:int>
-[__readNone]
-[ForceInline]
-[require(cpp_cuda_glsl_hlsl_spirv, GLSL_130)]
-public vector<T,N> asinh(vector<T,N> x)
-{
-    __target_switch
-    {
-    case glsl: __intrinsic_asm "asinh";
-    case spirv: return spirv_asm {
-        OpExtInst $$vector<T,N> result glsl450 Asinh $x
-    };
-    default:
-        VECTOR_MAP_UNARY(T, N, asinh, x);
-    }
-}
-
-__generic<T : __BuiltinFloatingPointType>
-[__readNone]
-[ForceInline]
-[require(cpp_cuda_glsl_hlsl_spirv, GLSL_130)]
-public T acosh(T x)
-{
-    __target_switch
-    {
-    case cpp: __intrinsic_asm "$P_acosh($0)";
-    case cuda: __intrinsic_asm "$P_acosh($0)";
-    case glsl: __intrinsic_asm "acosh";
-    case spirv: return spirv_asm {
-        OpExtInst $$T result glsl450 Acosh $x
-    };
-    default:
-        return log(x + sqrt( x * x - T(1)));
-    }
-}
-
-__generic<T : __BuiltinFloatingPointType, let N:int>
-[__readNone]
-[ForceInline]
-[require(cpp_cuda_glsl_hlsl_spirv, GLSL_130)]
-public vector<T,N> acosh(vector<T,N> x)
-{
-    __target_switch
-    {
-    case glsl: __intrinsic_asm "acosh";
-    case spirv: return spirv_asm {
-        OpExtInst $$vector<T,N> result glsl450 Acosh $x
-    };
-    default:
-        VECTOR_MAP_UNARY(T, N, acosh, x);
-    }
-}
-
-__generic<T : __BuiltinFloatingPointType>
-[__readNone]
-[ForceInline]
-[require(cpp_cuda_glsl_hlsl_spirv, GLSL_130)]
-public T atanh(T x)
-{
-    __target_switch
-    {
-    case cpp: __intrinsic_asm "$P_atanh($0)";
-    case cuda: __intrinsic_asm "$P_atanh($0)";
-    case glsl: __intrinsic_asm "atanh";
-    case spirv: return spirv_asm {
-        OpExtInst $$T result glsl450 Atanh $x
-    };
-    default:
-        return T(0.5) * log((T(1) + x) / (T(1) - x));
-    }
-}
-
-__generic<T : __BuiltinFloatingPointType, let N:int>
-[__readNone]
-[ForceInline]
-[require(cpp_cuda_glsl_hlsl_spirv, sm_2_0_GLSL_140)]
-public vector<T,N> atanh(vector<T,N> x)
-{
-    __target_switch
-    {
-    case glsl: __intrinsic_asm "atanh";
-    case spirv: return spirv_asm {
-        OpExtInst $$vector<T,N> result glsl450 Atanh $x
-    };
-    default:
-        VECTOR_MAP_UNARY(T, N, atanh, x);
-    }
-}
-
 //
 // Section 8.2. Exponential Functions
 //
@@ -458,66 +350,19 @@ public vector<T, N> inversesqrt(vector<T, N> x)
 __generic<T : __BuiltinFloatingPointType>
 [__readNone]
 [ForceInline]
-[require(cpp_cuda_glsl_hlsl_spirv, GLSL_130)]
+[require(cpp_cuda_glsl_hlsl_metal_spirv, GLSL_130)]
 public T roundEven(T x)
 {
-    __target_switch
-    {
-    case glsl: __intrinsic_asm "roundEven";
-    case spirv: return spirv_asm {
-        OpExtInst $$T result glsl450 RoundEven $x
-    };
-    default:
-        T nearest = round(x);
-
-        // Check if the value is exactly halfway between two integers
-        if (abs(x - nearest) == T(0.5))
-        {
-            // If halfway, choose the even number
-            if (mod(nearest, T(2)) != T(0))
-            {
-                // If the nearest number is odd,
-                // move to the closest even number
-                nearest -= ((x < nearest) ? T(1) : T(-1));
-            }
-        }
-        return nearest;
-    }
+    return rint(x);
 }
 
 __generic<T : __BuiltinFloatingPointType, let N:int>
 [__readNone]
 [ForceInline]
-[require(cpp_cuda_glsl_hlsl_spirv, GLSL_130)]
+[require(cpp_cuda_glsl_hlsl_metal_spirv, GLSL_130)]
 public vector<T,N> roundEven(vector<T,N> x)
 {
-    __target_switch
-    {
-    case glsl: __intrinsic_asm "roundEven";
-    case spirv: return spirv_asm {
-        OpExtInst $$vector<T,N> result glsl450 RoundEven $x
-    };
-    default:
-        VECTOR_MAP_UNARY(T, N, roundEven, x);
-    }
-}
-
-__generic<T : __BuiltinFloatingPointType>
-[__readNone]
-[ForceInline]
-[require(cpp_cuda_glsl_hlsl_spirv, sm_2_0_GLSL_140)]
-public T fract(T x)
-{
-    return frac(x);
-}
-
-__generic<T : __BuiltinFloatingPointType, let N:int>
-[__readNone]
-[ForceInline]
-[require(cpp_cuda_glsl_hlsl_spirv, sm_2_0_GLSL_140)]
-public vector<T, N> fract(vector<T, N> x)
-{
-    return frac(x);
+    return rint(x);
 }
 
 __generic<T : __BuiltinFloatingPointType>
@@ -824,44 +669,6 @@ uint float2half(float f)
     return (s | e | m);
 }
 
-__generic<T : __BuiltinFloatingPointType, E : __BuiltinIntegerType>
-[__readNone]
-[require(cpp_cuda_glsl_hlsl_spirv, sm_2_0_GLSL_140)]
-public T ldexp(T x, E exp)
-{
-    __target_switch
-    {
-    case hlsl: __intrinsic_asm "ldexp";
-    case glsl: __intrinsic_asm "ldexp";
-    case spirv: return spirv_asm {
-        OpExtInst $$T result glsl450 Ldexp $x $exp
-    };
-    default:
-        return ldexp(x, __floatCast<T>(exp));
-    }
-}
-
-__generic<T : __BuiltinFloatingPointType, E : __BuiltinIntegerType, let N : int>
-[__readNone]
-[require(cpp_cuda_glsl_hlsl_spirv, sm_2_0_GLSL_140)]
-public vector<T, N> ldexp(vector<T, N> x, vector<E, N> exp)
-{
-    __target_switch
-    {
-    case hlsl: __intrinsic_asm "ldexp";
-    case glsl: __intrinsic_asm "ldexp";
-    case spirv: return spirv_asm {
-        OpExtInst $$vector<T,N> result glsl450 Ldexp $x $exp
-    };
-    default:
-        vector<T,N> temp;
-        [ForceUnroll]
-        for (int i = 0; i < N; ++i)
-            temp[i] = __floatCast<T>(exp[i]);
-        return ldexp(x, temp);
-    }
-}
-
 [__readNone]
 [ForceInline]
 [require(cpp_cuda_glsl_hlsl_spirv, shader5_sm_4_0)]