From 305b62cf4f7e2a4ca3cc56109003aed6bde61c25 Mon Sep 17 00:00:00 2001 From: Mikko Rasa Date: Tue, 6 Apr 2021 12:54:06 +0300 Subject: [PATCH] Adjust GLSL builtins Remove some functions which are complicated to implement in SPIR-V due to the use of ad-hoc struct types or pointers. I may add them back later if there turns out to be some use for them. Add inline definitions for some overloads taking mixed vector and scalar arguments. SPIR-V operations require all operands to have the same number of elements. Builtin functions are always inlined even if used more than once. --- builtin_data/_builtin.glsl | 102 +++++++++++++++---------------------- scripts/builtin_funcs.py | 40 +++++++-------- source/glsl/optimize.cpp | 4 +- source/glsl/optimize.h | 2 +- 4 files changed, 62 insertions(+), 86 deletions(-) diff --git a/builtin_data/_builtin.glsl b/builtin_data/_builtin.glsl index aa36d5c6..124a117b 100644 --- a/builtin_data/_builtin.glsl +++ b/builtin_data/_builtin.glsl @@ -169,13 +169,9 @@ float mod(float x, float y); vec2 mod(vec2 x, vec2 y); vec3 mod(vec3 x, vec3 y); vec4 mod(vec4 x, vec4 y); -vec2 mod(vec2 x, float y); -vec3 mod(vec3 x, float y); -vec4 mod(vec4 x, float y); -float modf(float x, out float y); -vec2 modf(vec2 x, out vec2 y); -vec3 modf(vec3 x, out vec3 y); -vec4 modf(vec4 x, out vec4 y); +vec2 mod(vec2 x, float y) { return mod(x, vec2(y)); } +vec3 mod(vec3 x, float y) { return mod(x, vec3(y)); } +vec4 mod(vec4 x, float y) { return mod(x, vec4(y)); } float min(float x, float y); vec2 min(vec2 x, vec2 y); vec3 min(vec3 x, vec3 y); @@ -184,12 +180,12 @@ int min(int x, int y); ivec2 min(ivec2 x, ivec2 y); ivec3 min(ivec3 x, ivec3 y); ivec4 min(ivec4 x, ivec4 y); -vec2 min(vec2 x, float y); -vec3 min(vec3 x, float y); -vec4 min(vec4 x, float y); -ivec2 min(ivec2 x, int y); -ivec3 min(ivec3 x, int y); -ivec4 min(ivec4 x, int y); +vec2 min(vec2 x, float y) { return min(x, vec2(y)); } +vec3 min(vec3 x, float y) { return min(x, vec3(y)); } +vec4 min(vec4 x, float y) { return min(x, vec4(y)); } +ivec2 min(ivec2 x, int y) { return min(x, ivec2(y)); } +ivec3 min(ivec3 x, int y) { return min(x, ivec3(y)); } +ivec4 min(ivec4 x, int y) { return min(x, ivec4(y)); } float max(float x, float y); vec2 max(vec2 x, vec2 y); vec3 max(vec3 x, vec3 y); @@ -198,12 +194,12 @@ int max(int x, int y); ivec2 max(ivec2 x, ivec2 y); ivec3 max(ivec3 x, ivec3 y); ivec4 max(ivec4 x, ivec4 y); -vec2 max(vec2 x, float y); -vec3 max(vec3 x, float y); -vec4 max(vec4 x, float y); -ivec2 max(ivec2 x, int y); -ivec3 max(ivec3 x, int y); -ivec4 max(ivec4 x, int y); +vec2 max(vec2 x, float y) { return max(x, vec2(y)); } +vec3 max(vec3 x, float y) { return max(x, vec3(y)); } +vec4 max(vec4 x, float y) { return max(x, vec4(y)); } +ivec2 max(ivec2 x, int y) { return max(x, ivec2(y)); } +ivec3 max(ivec3 x, int y) { return max(x, ivec3(y)); } +ivec4 max(ivec4 x, int y) { return max(x, ivec4(y)); } float clamp(float x, float minVal, float maxVal); vec2 clamp(vec2 x, vec2 minVal, vec2 maxVal); vec3 clamp(vec3 x, vec3 minVal, vec3 maxVal); @@ -212,19 +208,20 @@ int clamp(int x, int minVal, int maxVal); ivec2 clamp(ivec2 x, ivec2 minVal, ivec2 maxVal); ivec3 clamp(ivec3 x, ivec3 minVal, ivec3 maxVal); ivec4 clamp(ivec4 x, ivec4 minVal, ivec4 maxVal); -vec2 clamp(vec2 x, float minVal, float maxVal); -vec3 clamp(vec3 x, float minVal, float maxVal); -vec4 clamp(vec4 x, float minVal, float maxVal); -ivec2 clamp(ivec2 x, int minVal, int maxVal); -ivec3 clamp(ivec3 x, int minVal, int maxVal); -ivec4 clamp(ivec4 x, int minVal, int maxVal); +vec2 clamp(vec2 x, float minVal, float maxVal) { return clamp(x, vec2(minVal), vec2(maxVal)); } +vec3 clamp(vec3 x, float minVal, float maxVal) { return clamp(x, vec3(minVal), vec3(maxVal)); } +vec4 clamp(vec4 x, float minVal, float maxVal) { return clamp(x, vec4(minVal), vec4(maxVal)); } +ivec2 clamp(ivec2 x, int minVal, int maxVal) { return clamp(x, ivec2(minVal), ivec2(maxVal)); } +ivec3 clamp(ivec3 x, int minVal, int maxVal) { return clamp(x, ivec3(minVal), ivec3(maxVal)); } +ivec4 clamp(ivec4 x, int minVal, int maxVal) { return clamp(x, ivec4(minVal), ivec4(maxVal)); } float mix(float x, float y, float a); vec2 mix(vec2 x, vec2 y, vec2 a); vec3 mix(vec3 x, vec3 y, vec3 a); vec4 mix(vec4 x, vec4 y, vec4 a); -vec2 mix(vec2 x, vec2 y, float a); -vec3 mix(vec3 x, vec3 y, float a); -vec4 mix(vec4 x, vec4 y, float a); +float mix(float x, float y, float a) { return mix(x, y, float(a)); } +vec2 mix(vec2 x, vec2 y, float a) { return mix(x, y, vec2(a)); } +vec3 mix(vec3 x, vec3 y, float a) { return mix(x, y, vec3(a)); } +vec4 mix(vec4 x, vec4 y, float a) { return mix(x, y, vec4(a)); } float mix(float x, float y, bool a); vec2 mix(vec2 x, vec2 y, bvec2 a); vec3 mix(vec3 x, vec3 y, bvec3 a); @@ -237,16 +234,16 @@ float step(float edge, float x); vec2 step(vec2 edge, vec2 x); vec3 step(vec3 edge, vec3 x); vec4 step(vec4 edge, vec4 x); -vec2 step(float edge, vec2 x); -vec3 step(float edge, vec3 x); -vec4 step(float edge, vec4 x); +vec2 step(float edge, vec2 x) { return step(vec2(edge), x); } +vec3 step(float edge, vec3 x) { return step(vec3(edge), x); } +vec4 step(float edge, vec4 x) { return step(vec4(edge), x); } float smoothstep(float edge0, float edge1, float x); vec2 smoothstep(vec2 edge0, vec2 edge1, vec2 x); vec3 smoothstep(vec3 edge0, vec3 edge1, vec3 x); vec4 smoothstep(vec4 edge0, vec4 edge1, vec4 x); -vec2 smoothstep(float edge0, float edge1, vec2 x); -vec3 smoothstep(float edge0, float edge1, vec3 x); -vec4 smoothstep(float edge0, float edge1, vec4 x); +vec2 smoothstep(float edge0, float edge1, vec2 x) { return smoothstep(vec2(edge0), vec2(edge1), x); } +vec3 smoothstep(float edge0, float edge1, vec3 x) { return smoothstep(vec3(edge0), vec3(edge1), x); } +vec4 smoothstep(float edge0, float edge1, vec4 x) { return smoothstep(vec4(edge0), vec4(edge1), x); } bool isnan(float x); bvec2 isnan(vec2 x); bvec3 isnan(vec3 x); @@ -255,26 +252,10 @@ bool isinf(float x); bvec2 isinf(vec2 x); bvec3 isinf(vec3 x); bvec4 isinf(vec4 x); -int floatBitsToInt(float value); -ivec2 floatBitsToInt(vec2 value); -ivec3 floatBitsToInt(vec3 value); -ivec4 floatBitsToInt(vec4 value); -float intBitsToFloat(int value); -vec2 intBitsToFloat(ivec2 value); -vec3 intBitsToFloat(ivec3 value); -vec4 intBitsToFloat(ivec4 value); float fma(float a, float b, float c); vec2 fma(vec2 a, vec2 b, vec2 c); vec3 fma(vec3 a, vec3 b, vec3 c); vec4 fma(vec4 a, vec4 b, vec4 c); -float frexp(float x, out int exp); -vec2 frexp(vec2 x, out ivec2 exp); -vec3 frexp(vec3 x, out ivec3 exp); -vec4 frexp(vec4 x, out ivec4 exp); -float ldexp(float x, int exp); -vec2 ldexp(vec2 x, ivec2 exp); -vec3 ldexp(vec3 x, ivec3 exp); -vec4 ldexp(vec4 x, ivec4 exp); float length(float x); float length(vec2 x); float length(vec3 x); @@ -296,14 +277,14 @@ float faceforward(float N, float I, float Nref); vec2 faceforward(vec2 N, vec2 I, vec2 Nref); vec3 faceforward(vec3 N, vec3 I, vec3 Nref); vec4 faceforward(vec4 N, vec4 I, vec4 Nref); -float reflect(float N, float I); -vec2 reflect(vec2 N, vec2 I); -vec3 reflect(vec3 N, vec3 I); -vec4 reflect(vec4 N, vec4 I); -float refract(float N, float I, float eta); -vec2 refract(vec2 N, vec2 I, float eta); -vec3 refract(vec3 N, vec3 I, float eta); -vec4 refract(vec4 N, vec4 I, float eta); +float reflect(float I, float N); +vec2 reflect(vec2 I, vec2 N); +vec3 reflect(vec3 I, vec3 N); +vec4 reflect(vec4 I, vec4 N); +float refract(float I, float N, float eta); +vec2 refract(vec2 I, vec2 N, float eta); +vec3 refract(vec3 I, vec3 N, float eta); +vec4 refract(vec4 I, vec4 N, float eta); mat2 matrixCompMult(mat2 x, mat2 y); mat3 matrixCompMult(mat3 x, mat3 y); mat4 matrixCompMult(mat4 x, mat4 y); @@ -382,10 +363,6 @@ bool all(bvec4 x); bool not(bvec2 x); bool not(bvec3 x); bool not(bvec4 x); -void imulExtended(int x, int y, out int msb, out int lsb); -void imulExtended(ivec2 x, ivec2 y, out ivec2 msb, out ivec2 lsb); -void imulExtended(ivec3 x, ivec3 y, out ivec3 msb, out ivec3 lsb); -void imulExtended(ivec4 x, ivec4 y, out ivec4 msb, out ivec4 lsb); int bitfieldExtract(int value, int offset, int bits); ivec2 bitfieldExtract(ivec2 value, int offset, int bits); ivec3 bitfieldExtract(ivec3 value, int offset, int bits); @@ -468,6 +445,7 @@ in gl_PerVertex float gl_ClipDistance[]; } gl_in[]; in int gl_PrimitiveIDIn; +in int gl_InvocationID; out gl_PerVertex { vec4 gl_Position; diff --git a/scripts/builtin_funcs.py b/scripts/builtin_funcs.py index e7618877..24cc3643 100755 --- a/scripts/builtin_funcs.py +++ b/scripts/builtin_funcs.py @@ -41,6 +41,7 @@ traits = { float32vectypes = ("vec2", "vec3", "vec4") float32types = ("float",)+float32vectypes +floatvectypes = float32vectypes floattypes = float32types int32vectypes = ("ivec2", "ivec3", "ivec4") int32types = ("int",)+int32vectypes @@ -91,28 +92,23 @@ shared_funcs = [ ("T ceil(T x)", floattypes), ("T fract(T x)", floattypes), ("T mod(T x, T y)", floattypes), - ("T mod(T x, T::Base y)", floattypes), - ("T modf(T x, out T y)", floattypes), + ("T mod(T x, T::Base y)", "mod(x, T(y))", floatvectypes), ("T min(T x, T y)", arithmetictypes), - ("T min(T x, T::Base y)", arithmetictypes), + ("T min(T x, T::Base y)", "min(x, T(y))", arithmeticvectypes), ("T max(T x, T y)", arithmetictypes), - ("T max(T x, T::Base y)", arithmetictypes), + ("T max(T x, T::Base y)", "max(x, T(y))", arithmeticvectypes), ("T clamp(T x, T minVal, T maxVal)", arithmetictypes), - ("T clamp(T x, T::Base minVal, T::Base maxVal)", arithmetictypes), + ("T clamp(T x, T::Base minVal, T::Base maxVal)", "clamp(x, T(minVal), T(maxVal))", arithmeticvectypes), ("T mix(T x, T y, T a)", floattypes), - ("T mix(T x, T y, T::Base a)", floattypes), + ("T mix(T x, T y, T::Base a)", "mix(x, y, T(a))", floattypes), ("T mix(T x, T y, bool[T::Dim] a)", arithmetictypes), ("T step(T edge, T x)", floattypes), - ("T step(T::Base edge, T x)", floattypes), + ("T step(T::Base edge, T x)", "step(T(edge), x)", floatvectypes), ("T smoothstep(T edge0, T edge1, T x)", floattypes), - ("T smoothstep(T::Base edge0, T::Base edge1, T x)", floattypes), + ("T smoothstep(T::Base edge0, T::Base edge1, T x)", "smoothstep(T(edge0), T(edge1), x)", floatvectypes), ("bool[T::Dim] isnan(T x)", floattypes), ("bool[T::Dim] isinf(T x)", floattypes), - ("int[T::Dim] floatBitsToInt(T value)", float32types), - ("T intBitsToFloat(int[T::Dim] value)", float32types), ("T fma(T a, T b, T c)", floattypes), - ("T frexp(T x, out int[T::Dim] exp)", floattypes), - ("T ldexp(T x, int[T::Dim] exp)", floattypes), # Geometric ("T::Base length(T x)", floattypes), @@ -121,8 +117,8 @@ shared_funcs = [ "vec3 cross(vec3 x)", ("T normalize(T x)", floattypes), ("T faceforward(T N, T I, T Nref)", floattypes), - ("T reflect(T N, T I)", floattypes), - ("T refract(T N, T I, float eta)", floattypes), + ("T reflect(T I, T N)", floattypes), + ("T refract(T I, T N, float eta)", floattypes), # Matrix ("T matrixCompMult(T x, T y)", matrixtypes), @@ -143,7 +139,6 @@ shared_funcs = [ ("bool not(T x)", boolvectypes), # Integer - ("void imulExtended(T x, T y, out T msb, out T lsb)", int32types), ("T bitfieldExtract(T value, int offset, int bits)", int32types), ("T bitfieldInsert(T value, T insert, int offset, int bits)", int32types), ("T bitfieldReverse(T value)", int32types), @@ -235,10 +230,9 @@ def expand_template(template, gentype): result += " " special = not t[0].isalpha() - if t==",": - result += ", " - else: - result += t + result += t + if t[-1]==",": + result += " " i += advance @@ -249,10 +243,14 @@ def generate_functions(funcs): generated = set() for f in funcs: if type(f)==tuple: - for t in f[1]: + for t in f[-1]: decl = expand_template(f[0], t) + if len(f)>=3: + decl += " {{ return {}; }}".format(expand_template(f[1], t)) + else: + decl += ";" if not decl in generated: - out_lines.append(decl+";\n") + out_lines.append(decl+"\n") generated.add(decl) else: out_lines.append(f+";\n") diff --git a/source/glsl/optimize.cpp b/source/glsl/optimize.cpp index 927047cc..0f765c2a 100644 --- a/source/glsl/optimize.cpp +++ b/source/glsl/optimize.cpp @@ -75,7 +75,7 @@ void InlineableFunctionLocator::visit(FunctionCall &call) ++count; /* Don't inline functions which are called more than once or are called recursively. */ - if(count>1 || def==current_function) + if((count>1 && def->source!=BUILTIN_SOURCE) || def==current_function) inlineable.erase(def); } @@ -89,7 +89,7 @@ void InlineableFunctionLocator::visit(FunctionDeclaration &func) has_out_params = ((*i)->interface=="out"); unsigned &count = refcounts[func.definition]; - if(count<=1 && !has_out_params) + if((count<=1 || func.source==BUILTIN_SOURCE) && !has_out_params) inlineable.insert(func.definition); SetForScope set(current_function, &func); diff --git a/source/glsl/optimize.h b/source/glsl/optimize.h index 249a859a..b7b39176 100644 --- a/source/glsl/optimize.h +++ b/source/glsl/optimize.h @@ -27,7 +27,7 @@ private: /** Finds functions which are candidates for inlining. Currently this means functions which have no flow control statements, no more than one return -statement, and are only called once. */ +statement, and are either builtins or only called once. */ class InlineableFunctionLocator: private TraversingVisitor { private: -- 2.45.2