| /* |
| * Copyright 2021 Google LLC |
| * |
| * Use of this source code is governed by a BSD-style license that can be |
| * found in the LICENSE file. |
| */ |
| |
| #include "src/sksl/SkSLAnalysis.h" |
| #include "src/sksl/SkSLConstantFolder.h" |
| #include "src/sksl/SkSLContext.h" |
| #include "src/sksl/SkSLProgramSettings.h" |
| #include "src/sksl/ir/SkSLChildCall.h" |
| #include "src/sksl/ir/SkSLConstructorCompound.h" |
| #include "src/sksl/ir/SkSLFunctionCall.h" |
| #include "src/sksl/ir/SkSLLiteral.h" |
| |
| #include "include/private/SkFloatingPoint.h" |
| #include "include/sksl/DSLCore.h" |
| #include "src/core/SkMatrixInvert.h" |
| |
| #include <array> |
| |
| namespace SkSL { |
| |
| using IntrinsicArguments = std::array<const Expression*, 3>; |
| |
| static bool has_compile_time_constant_arguments(const ExpressionArray& arguments) { |
| for (const std::unique_ptr<Expression>& arg : arguments) { |
| const Expression* expr = ConstantFolder::GetConstantValueForVariable(*arg); |
| if (!expr->isCompileTimeConstant()) { |
| return false; |
| } |
| } |
| return true; |
| } |
| |
| static double as_double(const Expression* expr) { |
| if (expr && expr->is<Literal>()) { |
| return expr->as<Literal>().value(); |
| } |
| return 0.0; |
| } |
| |
| template <typename T> |
| static void type_check_expression(const Expression& expr); |
| |
| template <> |
| void type_check_expression<float>(const Expression& expr) { |
| SkASSERT(expr.type().componentType().isFloat()); |
| } |
| |
| template <> |
| void type_check_expression<SKSL_INT>(const Expression& expr) { |
| SkASSERT(expr.type().componentType().isInteger()); |
| } |
| |
| template <> |
| void type_check_expression<bool>(const Expression& expr) { |
| SkASSERT(expr.type().componentType().isBoolean()); |
| } |
| |
| static std::unique_ptr<Expression> assemble_compound(const Context& context, |
| int offset, |
| const Type& returnType, |
| double value[]) { |
| int numSlots = returnType.slotCount(); |
| ExpressionArray array; |
| array.reserve_back(numSlots); |
| for (int index = 0; index < numSlots; ++index) { |
| array.push_back(Literal::Make(offset, value[index], &returnType.componentType())); |
| } |
| return ConstructorCompound::Make(context, offset, returnType, std::move(array)); |
| } |
| |
| using CoalesceFn = double (*)(double, double, double); |
| using FinalizeFn = double (*)(double); |
| |
| static std::unique_ptr<Expression> coalesce_n_way_vector(const Expression* arg0, |
| const Expression* arg1, |
| double startingState, |
| const Type& returnType, |
| CoalesceFn coalesce, |
| FinalizeFn finalize) { |
| // Takes up to two vector or scalar arguments and coalesces them in sequence: |
| // scalar = startingState; |
| // scalar = coalesce(scalar, arg0.x, arg1.x); |
| // scalar = coalesce(scalar, arg0.y, arg1.y); |
| // scalar = coalesce(scalar, arg0.z, arg1.z); |
| // scalar = coalesce(scalar, arg0.w, arg1.w); |
| // scalar = finalize(scalar); |
| // |
| // If an argument is null, zero is passed to the coalesce function. If the arguments are a mix |
| // of scalars and vectors, the scalars is interpreted as a vector containing the same value for |
| // every component. |
| |
| int offset = arg0->fOffset; |
| |
| const Type& vecType = arg0->type().isVector() ? arg0->type() : |
| (arg1 && arg1->type().isVector()) ? arg1->type() : |
| arg0->type(); |
| SkASSERT( arg0->type().componentType() == vecType.componentType()); |
| SkASSERT(!arg1 || arg1->type().componentType() == vecType.componentType()); |
| |
| double value = startingState; |
| int arg0Index = 0; |
| int arg1Index = 0; |
| for (int index = 0; index < vecType.columns(); ++index) { |
| const Expression* arg0Subexpr = arg0->getConstantSubexpression(arg0Index); |
| arg0Index += arg0->type().isVector() ? 1 : 0; |
| SkASSERT(arg0Subexpr); |
| |
| const Expression* arg1Subexpr = nullptr; |
| if (arg1) { |
| arg1Subexpr = arg1->getConstantSubexpression(arg1Index); |
| arg1Index += arg1->type().isVector() ? 1 : 0; |
| SkASSERT(arg1Subexpr); |
| } |
| |
| value = coalesce(value, as_double(arg0Subexpr), as_double(arg1Subexpr)); |
| |
| // If coalescing the intrinsic yields a non-finite value, do not optimize. |
| if (!std::isfinite(value)) { |
| return nullptr; |
| } |
| } |
| |
| if (finalize) { |
| value = finalize(value); |
| } |
| |
| return Literal::Make(offset, value, &returnType); |
| } |
| |
| template <typename T> |
| static std::unique_ptr<Expression> coalesce_vector(const IntrinsicArguments& arguments, |
| double startingState, |
| const Type& returnType, |
| CoalesceFn coalesce, |
| FinalizeFn finalize) { |
| SkASSERT(arguments[0]); |
| SkASSERT(!arguments[1]); |
| type_check_expression<T>(*arguments[0]); |
| |
| return coalesce_n_way_vector(arguments[0], /*arg1=*/nullptr, |
| startingState, returnType, coalesce, finalize); |
| } |
| |
| template <typename T> |
| static std::unique_ptr<Expression> coalesce_pairwise_vectors(const IntrinsicArguments& arguments, |
| double startingState, |
| const Type& returnType, |
| CoalesceFn coalesce, |
| FinalizeFn finalize) { |
| SkASSERT(arguments[0]); |
| SkASSERT(arguments[1]); |
| SkASSERT(!arguments[2]); |
| type_check_expression<T>(*arguments[0]); |
| type_check_expression<T>(*arguments[1]); |
| |
| return coalesce_n_way_vector(arguments[0], arguments[1], |
| startingState, returnType, coalesce, finalize); |
| } |
| |
| using CompareFn = bool (*)(double, double); |
| |
| static std::unique_ptr<Expression> optimize_comparison(const Context& context, |
| const IntrinsicArguments& arguments, |
| CompareFn compare) { |
| const Expression* left = arguments[0]; |
| const Expression* right = arguments[1]; |
| SkASSERT(left); |
| SkASSERT(right); |
| SkASSERT(!arguments[2]); |
| |
| const Type& type = left->type(); |
| SkASSERT(type.isVector()); |
| SkASSERT(type.componentType().isNumber()); |
| SkASSERT(type == right->type()); |
| |
| double array[4]; |
| |
| for (int index = 0; index < type.columns(); ++index) { |
| const Expression* leftSubexpr = left->getConstantSubexpression(index); |
| const Expression* rightSubexpr = right->getConstantSubexpression(index); |
| SkASSERT(leftSubexpr); |
| SkASSERT(rightSubexpr); |
| array[index] = compare(as_double(leftSubexpr), as_double(rightSubexpr)) ? 1.0 : 0.0; |
| } |
| |
| const Type& bvecType = context.fTypes.fBool->toCompound(context, type.columns(), /*rows=*/1); |
| return assemble_compound(context, left->fOffset, bvecType, array); |
| } |
| |
| using EvaluateFn = double (*)(double, double, double); |
| |
| static std::unique_ptr<Expression> evaluate_n_way_intrinsic(const Context& context, |
| const Expression* arg0, |
| const Expression* arg1, |
| const Expression* arg2, |
| const Type& returnType, |
| EvaluateFn eval) { |
| // Takes up to three arguments and evaluates all of them, left-to-right, in tandem. |
| // Equivalent to constructing a new compound value containing the results from: |
| // eval(arg0.x, arg1.x, arg2.x), |
| // eval(arg0.y, arg1.y, arg2.y), |
| // eval(arg0.z, arg1.z, arg2.z), |
| // eval(arg0.w, arg1.w, arg2.w) |
| // |
| // If an argument is null, zero is passed to the evaluation function. If the arguments are a mix |
| // of scalars and compounds, scalars are interpreted as a compound containing the same value for |
| // every component. |
| |
| int slots = returnType.slotCount(); |
| double array[16]; |
| |
| int arg0Index = 0; |
| int arg1Index = 0; |
| int arg2Index = 0; |
| for (int index = 0; index < slots; ++index) { |
| const Expression* arg0Subexpr = arg0->getConstantSubexpression(arg0Index); |
| arg0Index += arg0->type().isScalar() ? 0 : 1; |
| SkASSERT(arg0Subexpr); |
| |
| const Expression* arg1Subexpr = nullptr; |
| if (arg1) { |
| arg1Subexpr = arg1->getConstantSubexpression(arg1Index); |
| arg1Index += arg1->type().isScalar() ? 0 : 1; |
| SkASSERT(arg1Subexpr); |
| } |
| |
| const Expression* arg2Subexpr = nullptr; |
| if (arg2) { |
| arg2Subexpr = arg2->getConstantSubexpression(arg2Index); |
| arg2Index += arg2->type().isScalar() ? 0 : 1; |
| SkASSERT(arg2Subexpr); |
| } |
| |
| array[index] = eval(as_double(arg0Subexpr), as_double(arg1Subexpr), as_double(arg2Subexpr)); |
| |
| // If evaluation of the intrinsic yields a non-finite value, do not optimize. |
| if (!std::isfinite(array[index])) { |
| return nullptr; |
| } |
| } |
| |
| return assemble_compound(context, arg0->fOffset, returnType, array); |
| } |
| |
| template <typename T> |
| static std::unique_ptr<Expression> evaluate_intrinsic(const Context& context, |
| const IntrinsicArguments& arguments, |
| const Type& returnType, |
| EvaluateFn eval) { |
| SkASSERT(arguments[0]); |
| SkASSERT(!arguments[1]); |
| type_check_expression<T>(*arguments[0]); |
| |
| return evaluate_n_way_intrinsic(context, arguments[0], /*arg1=*/nullptr, /*arg2=*/nullptr, |
| returnType, eval); |
| } |
| |
| static std::unique_ptr<Expression> evaluate_intrinsic_numeric(const Context& context, |
| const IntrinsicArguments& arguments, |
| const Type& returnType, |
| EvaluateFn eval) { |
| SkASSERT(arguments[0]); |
| SkASSERT(!arguments[1]); |
| const Type& type = arguments[0]->type().componentType(); |
| |
| if (type.isFloat()) { |
| return evaluate_intrinsic<float>(context, arguments, returnType, eval); |
| } |
| if (type.isInteger()) { |
| return evaluate_intrinsic<SKSL_INT>(context, arguments, returnType, eval); |
| } |
| |
| SkDEBUGFAILF("unsupported type %s", type.description().c_str()); |
| return nullptr; |
| } |
| |
| static std::unique_ptr<Expression> evaluate_pairwise_intrinsic(const Context& context, |
| const IntrinsicArguments& arguments, |
| const Type& returnType, |
| EvaluateFn eval) { |
| SkASSERT(arguments[0]); |
| SkASSERT(arguments[1]); |
| SkASSERT(!arguments[2]); |
| const Type& type = arguments[0]->type().componentType(); |
| |
| if (type.isFloat()) { |
| type_check_expression<float>(*arguments[0]); |
| type_check_expression<float>(*arguments[1]); |
| } else if (type.isInteger()) { |
| type_check_expression<SKSL_INT>(*arguments[0]); |
| type_check_expression<SKSL_INT>(*arguments[1]); |
| } else { |
| SkDEBUGFAILF("unsupported type %s", type.description().c_str()); |
| return nullptr; |
| } |
| |
| return evaluate_n_way_intrinsic(context, arguments[0], arguments[1], /*arg2=*/nullptr, |
| returnType, eval); |
| } |
| |
| static std::unique_ptr<Expression> evaluate_3_way_intrinsic(const Context& context, |
| const IntrinsicArguments& arguments, |
| const Type& returnType, |
| EvaluateFn eval) { |
| SkASSERT(arguments[0]); |
| SkASSERT(arguments[1]); |
| SkASSERT(arguments[2]); |
| const Type& type = arguments[0]->type().componentType(); |
| |
| if (type.isFloat()) { |
| type_check_expression<float>(*arguments[0]); |
| type_check_expression<float>(*arguments[1]); |
| type_check_expression<float>(*arguments[2]); |
| } else if (type.isInteger()) { |
| type_check_expression<SKSL_INT>(*arguments[0]); |
| type_check_expression<SKSL_INT>(*arguments[1]); |
| type_check_expression<SKSL_INT>(*arguments[2]); |
| } else { |
| SkDEBUGFAILF("unsupported type %s", type.description().c_str()); |
| return nullptr; |
| } |
| |
| return evaluate_n_way_intrinsic(context, arguments[0], arguments[1], arguments[2], |
| returnType, eval); |
| } |
| |
| template <typename T1, typename T2> |
| static double pun_value(double val) { |
| // Interpret `val` as a value of type T1. |
| static_assert(sizeof(T1) == sizeof(T2)); |
| T1 inputValue = (T1)val; |
| // Reinterpret those bits as a value of type T2. |
| T2 outputValue; |
| memcpy(&outputValue, &inputValue, sizeof(T2)); |
| // Return the value-of-type-T2 as a double. (Non-finite values will prohibit optimization.) |
| return (double)outputValue; |
| } |
| |
| // Helper functions for optimizing all of our intrinsics. |
| namespace Intrinsics { |
| namespace { |
| |
| double coalesce_length(double a, double b, double) { return a + (b * b); } |
| double finalize_length(double a) { return std::sqrt(a); } |
| |
| double coalesce_distance(double a, double b, double c) { b -= c; return a + (b * b); } |
| double finalize_distance(double a) { return std::sqrt(a); } |
| |
| double coalesce_dot(double a, double b, double c) { return a + (b * c); } |
| double coalesce_any(double a, double b, double) { return a || b; } |
| double coalesce_all(double a, double b, double) { return a && b; } |
| |
| bool compare_lessThan(double a, double b) { return a < b; } |
| bool compare_lessThanEqual(double a, double b) { return a <= b; } |
| bool compare_greaterThan(double a, double b) { return a > b; } |
| bool compare_greaterThanEqual(double a, double b) { return a >= b; } |
| bool compare_equal(double a, double b) { return a == b; } |
| bool compare_notEqual(double a, double b) { return a != b; } |
| |
| double evaluate_radians(double a, double, double) { return a * 0.0174532925; } |
| double evaluate_degrees(double a, double, double) { return a * 57.2957795; } |
| double evaluate_sin(double a, double, double) { return std::sin(a); } |
| double evaluate_cos(double a, double, double) { return std::cos(a); } |
| double evaluate_tan(double a, double, double) { return std::tan(a); } |
| double evaluate_asin(double a, double, double) { return std::asin(a); } |
| double evaluate_acos(double a, double, double) { return std::acos(a); } |
| double evaluate_atan(double a, double, double) { return std::atan(a); } |
| double evaluate_atan2(double a, double b, double) { return std::atan2(a, b); } |
| double evaluate_asinh(double a, double, double) { return std::asinh(a); } |
| double evaluate_acosh(double a, double, double) { return std::acosh(a); } |
| double evaluate_atanh(double a, double, double) { return std::atanh(a); } |
| |
| double evaluate_pow(double a, double b, double) { return std::pow(a, b); } |
| double evaluate_exp(double a, double, double) { return std::exp(a); } |
| double evaluate_log(double a, double, double) { return std::log(a); } |
| double evaluate_exp2(double a, double, double) { return std::exp2(a); } |
| double evaluate_log2(double a, double, double) { return std::log2(a); } |
| double evaluate_sqrt(double a, double, double) { return std::sqrt(a); } |
| double evaluate_inversesqrt(double a, double, double) { |
| return sk_ieee_double_divide(1.0, std::sqrt(a)); |
| } |
| |
| double evaluate_abs(double a, double, double) { return std::abs(a); } |
| double evaluate_sign(double a, double, double) { return (a > 0) - (a < 0); } |
| double evaluate_floor(double a, double, double) { return std::floor(a); } |
| double evaluate_ceil(double a, double, double) { return std::ceil(a); } |
| double evaluate_fract(double a, double, double) { return a - std::floor(a); } |
| double evaluate_min(double a, double b, double) { return (a < b) ? a : b; } |
| double evaluate_max(double a, double b, double) { return (a > b) ? a : b; } |
| double evaluate_clamp(double x, double l, double h) { return (x < l) ? l : (x > h) ? h : x; } |
| double evaluate_saturate(double a, double, double) { return (a < 0) ? 0 : (a > 1) ? 1 : a; } |
| double evaluate_mix(double x, double y, double a) { return x * (1 - a) + y * a; } |
| double evaluate_step(double e, double x, double) { return (x < e) ? 0 : 1; } |
| double evaluate_mod(double a, double b, double) { |
| return a - b * std::floor(sk_ieee_double_divide(a, b)); |
| } |
| double evaluate_smoothstep(double edge0, double edge1, double x) { |
| double t = sk_ieee_double_divide(x - edge0, edge1 - edge0); |
| t = (t < 0) ? 0 : (t > 1) ? 1 : t; |
| return t * t * (3.0 - 2.0 * t); |
| } |
| |
| double evaluate_matrixCompMult(double x, double y, double) { return x * y; } |
| |
| double evaluate_not(double a, double, double) { return !a; } |
| double evaluate_sinh(double a, double, double) { return std::sinh(a); } |
| double evaluate_cosh(double a, double, double) { return std::cosh(a); } |
| double evaluate_tanh(double a, double, double) { return std::tanh(a); } |
| double evaluate_trunc(double a, double, double) { return std::trunc(a); } |
| double evaluate_round(double a, double, double) { |
| // The semantics of std::remainder guarantee a rounded-to-even result here, regardless of the |
| // current float-rounding mode. |
| return a - std::remainder(a, 1.0); |
| } |
| double evaluate_floatBitsToInt(double a, double, double) { return pun_value<float, int32_t> (a); } |
| double evaluate_floatBitsToUint(double a, double, double) { return pun_value<float, uint32_t>(a); } |
| double evaluate_intBitsToFloat(double a, double, double) { return pun_value<int32_t, float>(a); } |
| double evaluate_uintBitsToFloat(double a, double, double) { return pun_value<uint32_t, float>(a); } |
| |
| } // namespace |
| } // namespace Intrinsics |
| |
| static void extract_matrix(const Expression* expr, float mat[16]) { |
| size_t numSlots = expr->type().slotCount(); |
| for (size_t index = 0; index < numSlots; ++index) { |
| mat[index] = expr->getConstantSubexpression(index)->as<Literal>().floatValue(); |
| } |
| } |
| |
| static std::unique_ptr<Expression> optimize_intrinsic_call(const Context& context, |
| IntrinsicKind intrinsic, |
| const ExpressionArray& argArray, |
| const Type& returnType) { |
| // Replace constant variables with their literal values. |
| IntrinsicArguments arguments = {}; |
| SkASSERT(argArray.size() <= arguments.size()); |
| for (int index = 0; index < argArray.count(); ++index) { |
| arguments[index] = ConstantFolder::GetConstantValueForVariable(*argArray[index]); |
| } |
| |
| auto Get = [&](int idx, int col) -> float { |
| return arguments[idx]->getConstantSubexpression(col)->as<Literal>().floatValue(); |
| }; |
| |
| using namespace SkSL::dsl; |
| switch (intrinsic) { |
| // 8.1 : Angle and Trigonometry Functions |
| case k_radians_IntrinsicKind: |
| return evaluate_intrinsic<float>(context, arguments, returnType, |
| Intrinsics::evaluate_radians); |
| case k_degrees_IntrinsicKind: |
| return evaluate_intrinsic<float>(context, arguments, returnType, |
| Intrinsics::evaluate_degrees); |
| case k_sin_IntrinsicKind: |
| return evaluate_intrinsic<float>(context, arguments, returnType, |
| Intrinsics::evaluate_sin); |
| case k_cos_IntrinsicKind: |
| return evaluate_intrinsic<float>(context, arguments, returnType, |
| Intrinsics::evaluate_cos); |
| case k_tan_IntrinsicKind: |
| return evaluate_intrinsic<float>(context, arguments, returnType, |
| Intrinsics::evaluate_tan); |
| case k_sinh_IntrinsicKind: |
| return evaluate_intrinsic<float>(context, arguments, returnType, |
| Intrinsics::evaluate_sinh); |
| case k_cosh_IntrinsicKind: |
| return evaluate_intrinsic<float>(context, arguments, returnType, |
| Intrinsics::evaluate_cosh); |
| case k_tanh_IntrinsicKind: |
| return evaluate_intrinsic<float>(context, arguments, returnType, |
| Intrinsics::evaluate_tanh); |
| case k_asin_IntrinsicKind: |
| return evaluate_intrinsic<float>(context, arguments, returnType, |
| Intrinsics::evaluate_asin); |
| case k_acos_IntrinsicKind: |
| return evaluate_intrinsic<float>(context, arguments, returnType, |
| Intrinsics::evaluate_acos); |
| case k_atan_IntrinsicKind: |
| if (argArray.size() == 1) { |
| return evaluate_intrinsic<float>(context, arguments, returnType, |
| Intrinsics::evaluate_atan); |
| } else { |
| return evaluate_pairwise_intrinsic(context, arguments, returnType, |
| Intrinsics::evaluate_atan2); |
| } |
| case k_asinh_IntrinsicKind: |
| return evaluate_intrinsic<float>(context, arguments, returnType, |
| Intrinsics::evaluate_asinh); |
| |
| case k_acosh_IntrinsicKind: |
| return evaluate_intrinsic<float>(context, arguments, returnType, |
| Intrinsics::evaluate_acosh); |
| case k_atanh_IntrinsicKind: |
| return evaluate_intrinsic<float>(context, arguments, returnType, |
| Intrinsics::evaluate_atanh); |
| // 8.2 : Exponential Functions |
| case k_pow_IntrinsicKind: |
| return evaluate_pairwise_intrinsic(context, arguments, returnType, |
| Intrinsics::evaluate_pow); |
| case k_exp_IntrinsicKind: |
| return evaluate_intrinsic<float>(context, arguments, returnType, |
| Intrinsics::evaluate_exp); |
| case k_log_IntrinsicKind: |
| return evaluate_intrinsic<float>(context, arguments, returnType, |
| Intrinsics::evaluate_log); |
| case k_exp2_IntrinsicKind: |
| return evaluate_intrinsic<float>(context, arguments, returnType, |
| Intrinsics::evaluate_exp2); |
| case k_log2_IntrinsicKind: |
| return evaluate_intrinsic<float>(context, arguments, returnType, |
| Intrinsics::evaluate_log2); |
| case k_sqrt_IntrinsicKind: |
| return evaluate_intrinsic<float>(context, arguments, returnType, |
| Intrinsics::evaluate_sqrt); |
| case k_inversesqrt_IntrinsicKind: |
| return evaluate_intrinsic<float>(context, arguments, returnType, |
| Intrinsics::evaluate_inversesqrt); |
| // 8.3 : Common Functions |
| case k_abs_IntrinsicKind: |
| return evaluate_intrinsic_numeric(context, arguments, returnType, |
| Intrinsics::evaluate_abs); |
| case k_sign_IntrinsicKind: |
| return evaluate_intrinsic_numeric(context, arguments, returnType, |
| Intrinsics::evaluate_sign); |
| case k_floor_IntrinsicKind: |
| return evaluate_intrinsic<float>(context, arguments, returnType, |
| Intrinsics::evaluate_floor); |
| case k_ceil_IntrinsicKind: |
| return evaluate_intrinsic<float>(context, arguments, returnType, |
| Intrinsics::evaluate_ceil); |
| case k_fract_IntrinsicKind: |
| return evaluate_intrinsic<float>(context, arguments, returnType, |
| Intrinsics::evaluate_fract); |
| case k_mod_IntrinsicKind: |
| return evaluate_pairwise_intrinsic(context, arguments, returnType, |
| Intrinsics::evaluate_mod); |
| case k_min_IntrinsicKind: |
| return evaluate_pairwise_intrinsic(context, arguments, returnType, |
| Intrinsics::evaluate_min); |
| case k_max_IntrinsicKind: |
| return evaluate_pairwise_intrinsic(context, arguments, returnType, |
| Intrinsics::evaluate_max); |
| case k_clamp_IntrinsicKind: |
| return evaluate_3_way_intrinsic(context, arguments, returnType, |
| Intrinsics::evaluate_clamp); |
| case k_saturate_IntrinsicKind: |
| return evaluate_intrinsic<float>(context, arguments, returnType, |
| Intrinsics::evaluate_saturate); |
| case k_mix_IntrinsicKind: |
| if (arguments[2]->type().componentType().isBoolean()) { |
| const SkSL::Type& numericType = arguments[0]->type().componentType(); |
| |
| if (numericType.isFloat()) { |
| type_check_expression<float>(*arguments[0]); |
| type_check_expression<float>(*arguments[1]); |
| } else if (numericType.isInteger()) { |
| type_check_expression<SKSL_INT>(*arguments[0]); |
| type_check_expression<SKSL_INT>(*arguments[1]); |
| } else if (numericType.isBoolean()) { |
| type_check_expression<bool>(*arguments[0]); |
| type_check_expression<bool>(*arguments[1]); |
| } else { |
| SkDEBUGFAILF("unsupported type %s", numericType.description().c_str()); |
| return nullptr; |
| } |
| return evaluate_n_way_intrinsic(context, arguments[0], arguments[1], arguments[2], |
| returnType, Intrinsics::evaluate_mix); |
| } else { |
| return evaluate_3_way_intrinsic(context, arguments, returnType, |
| Intrinsics::evaluate_mix); |
| } |
| case k_step_IntrinsicKind: |
| return evaluate_pairwise_intrinsic(context, arguments, returnType, |
| Intrinsics::evaluate_step); |
| case k_smoothstep_IntrinsicKind: |
| return evaluate_3_way_intrinsic(context, arguments, returnType, |
| Intrinsics::evaluate_smoothstep); |
| case k_trunc_IntrinsicKind: |
| return evaluate_intrinsic<float>(context, arguments, returnType, |
| Intrinsics::evaluate_trunc); |
| case k_round_IntrinsicKind: // GLSL `round` documents its rounding mode as unspecified |
| case k_roundEven_IntrinsicKind: // and is allowed to behave identically to `roundEven`. |
| return evaluate_intrinsic<float>(context, arguments, returnType, |
| Intrinsics::evaluate_round); |
| case k_floatBitsToInt_IntrinsicKind: |
| return evaluate_intrinsic<float>(context, arguments, returnType, |
| Intrinsics::evaluate_floatBitsToInt); |
| case k_floatBitsToUint_IntrinsicKind: |
| return evaluate_intrinsic<float>(context, arguments, returnType, |
| Intrinsics::evaluate_floatBitsToUint); |
| case k_intBitsToFloat_IntrinsicKind: |
| return evaluate_intrinsic<SKSL_INT>(context, arguments, returnType, |
| Intrinsics::evaluate_intBitsToFloat); |
| case k_uintBitsToFloat_IntrinsicKind: |
| return evaluate_intrinsic<SKSL_INT>(context, arguments, returnType, |
| Intrinsics::evaluate_uintBitsToFloat); |
| // 8.4 : Floating-Point Pack and Unpack Functions |
| case k_packUnorm2x16_IntrinsicKind: { |
| auto Pack = [&](int n) -> unsigned int { |
| float x = Get(0, n); |
| return (int)std::round(Intrinsics::evaluate_clamp(x, 0.0, 1.0) * 65535.0); |
| }; |
| return UInt(((Pack(0) << 0) & 0x0000FFFF) | |
| ((Pack(1) << 16) & 0xFFFF0000)).release(); |
| } |
| case k_unpackUnorm2x16_IntrinsicKind: { |
| SKSL_INT x = arguments[0]->getConstantSubexpression(0)->as<Literal>().intValue(); |
| return Float2(double((x >> 0) & 0x0000FFFF) / 65535.0, |
| double((x >> 16) & 0x0000FFFF) / 65535.0).release(); |
| } |
| // 8.5 : Geometric Functions |
| case k_length_IntrinsicKind: |
| return coalesce_vector<float>(arguments, /*startingState=*/0, returnType, |
| Intrinsics::coalesce_length, |
| Intrinsics::finalize_length); |
| case k_distance_IntrinsicKind: |
| return coalesce_pairwise_vectors<float>(arguments, /*startingState=*/0, returnType, |
| Intrinsics::coalesce_distance, |
| Intrinsics::finalize_distance); |
| case k_dot_IntrinsicKind: |
| return coalesce_pairwise_vectors<float>(arguments, /*startingState=*/0, returnType, |
| Intrinsics::coalesce_dot, |
| /*finalize=*/nullptr); |
| case k_cross_IntrinsicKind: { |
| auto X = [&](int n) -> float { return Get(0, n); }; |
| auto Y = [&](int n) -> float { return Get(1, n); }; |
| SkASSERT(arguments[0]->type().columns() == 3); // the vec2 form is not a real intrinsic |
| |
| double vec[3] = {X(1) * Y(2) - Y(1) * X(2), |
| X(2) * Y(0) - Y(2) * X(0), |
| X(0) * Y(1) - Y(0) * X(1)}; |
| return assemble_compound(context, arguments[0]->fOffset, returnType, vec); |
| } |
| case k_normalize_IntrinsicKind: { |
| auto Vec = [&] { return DSLExpression{arguments[0]->clone()}; }; |
| return (Vec() / Length(Vec())).release(); |
| } |
| case k_faceforward_IntrinsicKind: { |
| auto N = [&] { return DSLExpression{arguments[0]->clone()}; }; |
| auto I = [&] { return DSLExpression{arguments[1]->clone()}; }; |
| auto NRef = [&] { return DSLExpression{arguments[2]->clone()}; }; |
| return (N() * Select(Dot(NRef(), I()) < 0, 1, -1)).release(); |
| } |
| case k_reflect_IntrinsicKind: { |
| auto I = [&] { return DSLExpression{arguments[0]->clone()}; }; |
| auto N = [&] { return DSLExpression{arguments[1]->clone()}; }; |
| return (I() - 2.0 * Dot(N(), I()) * N()).release(); |
| } |
| case k_refract_IntrinsicKind: { |
| auto I = [&] { return DSLExpression{arguments[0]->clone()}; }; |
| auto N = [&] { return DSLExpression{arguments[1]->clone()}; }; |
| auto Eta = [&] { return DSLExpression{arguments[2]->clone()}; }; |
| |
| std::unique_ptr<Expression> k = |
| (1 - Pow(Eta(), 2) * (1 - Pow(Dot(N(), I()), 2))).release(); |
| if (!k->is<Literal>()) { |
| return nullptr; |
| } |
| double kValue = k->as<Literal>().value(); |
| return ((kValue < 0) ? |
| (0 * I()) : |
| (Eta() * I() - (Eta() * Dot(N(), I()) + std::sqrt(kValue)) * N())).release(); |
| } |
| |
| // 8.6 : Matrix Functions |
| case k_matrixCompMult_IntrinsicKind: |
| return evaluate_pairwise_intrinsic(context, arguments, returnType, |
| Intrinsics::evaluate_matrixCompMult); |
| case k_transpose_IntrinsicKind: { |
| double mat[16]; |
| int index = 0; |
| for (int c = 0; c < returnType.columns(); ++c) { |
| for (int r = 0; r < returnType.rows(); ++r) { |
| mat[index++] = Get(0, (returnType.columns() * r) + c); |
| } |
| } |
| return assemble_compound(context, arguments[0]->fOffset, returnType, mat); |
| } |
| case k_outerProduct_IntrinsicKind: { |
| double mat[16]; |
| int index = 0; |
| for (int c = 0; c < returnType.columns(); ++c) { |
| for (int r = 0; r < returnType.rows(); ++r) { |
| mat[index++] = Get(0, r) * Get(1, c); |
| } |
| } |
| return assemble_compound(context, arguments[0]->fOffset, returnType, mat); |
| } |
| case k_determinant_IntrinsicKind: { |
| float mat[16]; |
| extract_matrix(arguments[0], mat); |
| float determinant; |
| switch (arguments[0]->type().slotCount()) { |
| case 4: |
| determinant = SkInvert2x2Matrix(mat, /*outMatrix=*/nullptr); |
| break; |
| case 9: |
| determinant = SkInvert3x3Matrix(mat, /*outMatrix=*/nullptr); |
| break; |
| case 16: |
| determinant = SkInvert4x4Matrix(mat, /*outMatrix=*/nullptr); |
| break; |
| default: |
| SkDEBUGFAILF("unsupported type %s", arguments[0]->type().description().c_str()); |
| return nullptr; |
| } |
| return Literal::MakeFloat(arguments[0]->fOffset, determinant, &returnType); |
| } |
| case k_inverse_IntrinsicKind: { |
| float mat[16] = {}; |
| extract_matrix(arguments[0], mat); |
| switch (arguments[0]->type().slotCount()) { |
| case 4: |
| if (SkInvert2x2Matrix(mat, mat) == 0.0f) { |
| return nullptr; |
| } |
| break; |
| case 9: |
| if (SkInvert3x3Matrix(mat, mat) == 0.0f) { |
| return nullptr; |
| } |
| break; |
| case 16: |
| if (SkInvert4x4Matrix(mat, mat) == 0.0f) { |
| return nullptr; |
| } |
| break; |
| default: |
| SkDEBUGFAILF("unsupported type %s", arguments[0]->type().description().c_str()); |
| return nullptr; |
| } |
| |
| double dmat[16]; |
| std::copy(mat, mat + SK_ARRAY_COUNT(mat), dmat); |
| return assemble_compound(context, arguments[0]->fOffset, returnType, dmat); |
| } |
| // 8.7 : Vector Relational Functions |
| case k_lessThan_IntrinsicKind: |
| return optimize_comparison(context, arguments, Intrinsics::compare_lessThan); |
| |
| case k_lessThanEqual_IntrinsicKind: |
| return optimize_comparison(context, arguments, Intrinsics::compare_lessThanEqual); |
| |
| case k_greaterThan_IntrinsicKind: |
| return optimize_comparison(context, arguments, Intrinsics::compare_greaterThan); |
| |
| case k_greaterThanEqual_IntrinsicKind: |
| return optimize_comparison(context, arguments, Intrinsics::compare_greaterThanEqual); |
| |
| case k_equal_IntrinsicKind: |
| return optimize_comparison(context, arguments, Intrinsics::compare_equal); |
| |
| case k_notEqual_IntrinsicKind: |
| return optimize_comparison(context, arguments, Intrinsics::compare_notEqual); |
| |
| case k_any_IntrinsicKind: |
| return coalesce_vector<bool>(arguments, /*startingState=*/false, returnType, |
| Intrinsics::coalesce_any, |
| /*finalize=*/nullptr); |
| case k_all_IntrinsicKind: |
| return coalesce_vector<bool>(arguments, /*startingState=*/true, returnType, |
| Intrinsics::coalesce_all, |
| /*finalize=*/nullptr); |
| case k_not_IntrinsicKind: |
| return evaluate_intrinsic<bool>(context, arguments, returnType, |
| Intrinsics::evaluate_not); |
| default: |
| return nullptr; |
| } |
| } |
| |
| bool FunctionCall::hasProperty(Property property) const { |
| if (property == Property::kSideEffects && |
| (this->function().modifiers().fFlags & Modifiers::kHasSideEffects_Flag)) { |
| return true; |
| } |
| for (const auto& arg : this->arguments()) { |
| if (arg->hasProperty(property)) { |
| return true; |
| } |
| } |
| return false; |
| } |
| |
| std::unique_ptr<Expression> FunctionCall::clone() const { |
| ExpressionArray cloned; |
| cloned.reserve_back(this->arguments().size()); |
| for (const std::unique_ptr<Expression>& arg : this->arguments()) { |
| cloned.push_back(arg->clone()); |
| } |
| return std::make_unique<FunctionCall>( |
| fOffset, &this->type(), &this->function(), std::move(cloned)); |
| } |
| |
| String FunctionCall::description() const { |
| String result = String(this->function().name()) + "("; |
| String separator; |
| for (const std::unique_ptr<Expression>& arg : this->arguments()) { |
| result += separator; |
| result += arg->description(); |
| separator = ", "; |
| } |
| result += ")"; |
| return result; |
| } |
| |
| std::unique_ptr<Expression> FunctionCall::Convert(const Context& context, |
| int offset, |
| const FunctionDeclaration& function, |
| ExpressionArray arguments) { |
| // Reject ES3 function calls in strict ES2 mode. |
| if (context.fConfig->strictES2Mode() && (function.modifiers().fFlags & Modifiers::kES3_Flag)) { |
| context.fErrors->error(offset, "call to '" + function.description() + "' is not supported"); |
| return nullptr; |
| } |
| |
| // Reject function calls with the wrong number of arguments. |
| if (function.parameters().size() != arguments.size()) { |
| String msg = "call to '" + function.name() + "' expected " + |
| to_string((int)function.parameters().size()) + " argument"; |
| if (function.parameters().size() != 1) { |
| msg += "s"; |
| } |
| msg += ", but found " + to_string(arguments.count()); |
| context.fErrors->error(offset, msg); |
| return nullptr; |
| } |
| |
| // Resolve generic types. |
| FunctionDeclaration::ParamTypes types; |
| const Type* returnType; |
| if (!function.determineFinalTypes(arguments, &types, &returnType)) { |
| String msg = "no match for " + function.name() + "("; |
| String separator; |
| for (const std::unique_ptr<Expression>& arg : arguments) { |
| msg += separator; |
| msg += arg->type().displayName(); |
| separator = ", "; |
| } |
| msg += ")"; |
| context.fErrors->error(offset, msg); |
| return nullptr; |
| } |
| |
| for (size_t i = 0; i < arguments.size(); i++) { |
| // Coerce each argument to the proper type. |
| arguments[i] = types[i]->coerceExpression(std::move(arguments[i]), context); |
| if (!arguments[i]) { |
| return nullptr; |
| } |
| // Update the refKind on out-parameters, and ensure that they are actually assignable. |
| const Modifiers& paramModifiers = function.parameters()[i]->modifiers(); |
| if (paramModifiers.fFlags & Modifiers::kOut_Flag) { |
| const VariableRefKind refKind = paramModifiers.fFlags & Modifiers::kIn_Flag |
| ? VariableReference::RefKind::kReadWrite |
| : VariableReference::RefKind::kPointer; |
| if (!Analysis::UpdateVariableRefKind(arguments[i].get(), refKind, context.fErrors)) { |
| return nullptr; |
| } |
| } |
| } |
| |
| if (function.intrinsicKind() == k_eval_IntrinsicKind) { |
| // This is a method call on an effect child. Translate it into a ChildCall, which simplifies |
| // handling in the generators and analysis code. |
| const Variable& child = *arguments.back()->as<VariableReference>().variable(); |
| arguments.pop_back(); |
| return ChildCall::Make(context, offset, returnType, child, std::move(arguments)); |
| } |
| |
| return Make(context, offset, returnType, function, std::move(arguments)); |
| } |
| |
| std::unique_ptr<Expression> FunctionCall::Make(const Context& context, |
| int offset, |
| const Type* returnType, |
| const FunctionDeclaration& function, |
| ExpressionArray arguments) { |
| SkASSERT(function.parameters().size() == arguments.size()); |
| |
| if (context.fConfig->fSettings.fOptimize) { |
| // We might be able to optimize built-in intrinsics. |
| if (function.isIntrinsic() && has_compile_time_constant_arguments(arguments)) { |
| // The function is an intrinsic and all inputs are compile-time constants. Optimize it. |
| if (std::unique_ptr<Expression> expr = optimize_intrinsic_call(context, |
| function.intrinsicKind(), |
| arguments, |
| *returnType)) { |
| return expr; |
| } |
| } |
| } |
| |
| return std::make_unique<FunctionCall>(offset, returnType, &function, std::move(arguments)); |
| } |
| |
| } // namespace SkSL |