From 81fce29d6e1f0a83e8a4170c7f24cdd93869d55a Mon Sep 17 00:00:00 2001 From: Thomas Lively Date: Wed, 5 May 2021 13:46:45 -0700 Subject: [PATCH] [WebAssembly] Add SIMD const_splat intrinsics These intrinsics do not correspond to their own underlying instruction, but are a convenience for the common case of materializing a constant vector that has the same value in each lane. Differential Revision: https://reviews.llvm.org/D101885 --- clang/lib/Headers/wasm_simd128.h | 37 ++++++++++++++++++++++--------- clang/test/Headers/wasm.c | 48 ++++++++++++++++++++++++++++++++++++++++ 2 files changed, 74 insertions(+), 11 deletions(-) diff --git a/clang/lib/Headers/wasm_simd128.h b/clang/lib/Headers/wasm_simd128.h index 2157125..e6fb0496 100644 --- a/clang/lib/Headers/wasm_simd128.h +++ b/clang/lib/Headers/wasm_simd128.h @@ -229,6 +229,11 @@ static __inline__ v128_t __DEFAULT_FN_ATTRS wasm_i32x4_make(int32_t __c0, return (v128_t)(__i32x4){__c0, __c1, __c2, __c3}; } +static __inline__ v128_t __DEFAULT_FN_ATTRS wasm_i64x2_make(int64_t __c0, + int64_t __c1) { + return (v128_t)(__i64x2){__c0, __c1}; +} + static __inline__ v128_t __DEFAULT_FN_ATTRS wasm_f32x4_make(float __c0, float __c1, float __c2, @@ -236,11 +241,6 @@ static __inline__ v128_t __DEFAULT_FN_ATTRS wasm_f32x4_make(float __c0, return (v128_t)(__f32x4){__c0, __c1, __c2, __c3}; } -static __inline__ v128_t __DEFAULT_FN_ATTRS wasm_i64x2_make(int64_t __c0, - int64_t __c1) { - return (v128_t)(__i64x2){__c0, __c1}; -} - static __inline__ v128_t __DEFAULT_FN_ATTRS wasm_f64x2_make(double __c0, double __c1) { return (v128_t)(__f64x2){__c0, __c1}; @@ -291,20 +291,20 @@ static __inline__ v128_t __DEFAULT_FN_ATTRS wasm_f64x2_make(double __c0, (v128_t)(__i32x4){__c0, __c1, __c2, __c3}; \ }) -#define wasm_f32x4_const(__c0, __c1, __c2, __c3) \ +#define wasm_i64x2_const(__c0, __c1) \ __extension__({ \ __REQUIRE_CONSTANT(__c0); \ __REQUIRE_CONSTANT(__c1); \ - __REQUIRE_CONSTANT(__c2); \ - __REQUIRE_CONSTANT(__c3); \ - (v128_t)(__f32x4){__c0, __c1, __c2, __c3}; \ + (v128_t)(__i64x2){__c0, __c1}; \ }) -#define wasm_i64x2_const(__c0, __c1) \ +#define wasm_f32x4_const(__c0, __c1, __c2, __c3) \ __extension__({ \ __REQUIRE_CONSTANT(__c0); \ __REQUIRE_CONSTANT(__c1); \ - (v128_t)(__i64x2){__c0, __c1}; \ + __REQUIRE_CONSTANT(__c2); \ + __REQUIRE_CONSTANT(__c3); \ + (v128_t)(__f32x4){__c0, __c1, __c2, __c3}; \ }) #define wasm_f64x2_const(__c0, __c1) \ @@ -314,6 +314,21 @@ static __inline__ v128_t __DEFAULT_FN_ATTRS wasm_f64x2_make(double __c0, (v128_t)(__f64x2){__c0, __c1}; \ }) +#define wasm_i8x16_const_splat(__c) \ + wasm_i8x16_const(__c, __c, __c, __c, __c, __c, __c, __c, __c, __c, __c, __c, \ + __c, __c, __c, __c) + +#define wasm_i16x8_const_splat(__c) \ + wasm_i16x8_const(__c, __c, __c, __c, __c, __c, __c, __c) + +#define wasm_i32x4_const_splat(__c) wasm_i32x4_const(__c, __c, __c, __c) + +#define wasm_i64x2_const_splat(__c) wasm_i64x2_const(__c, __c) + +#define wasm_f32x4_const_splat(__c) wasm_i32x4_const(__c, __c, __c, __c) + +#define wasm_f64x2_const_splat(__c) wasm_i64x2_const(__c, __c) + static __inline__ v128_t __DEFAULT_FN_ATTRS wasm_i8x16_splat(int8_t __a) { return (v128_t)(__i8x16){__a, __a, __a, __a, __a, __a, __a, __a, __a, __a, __a, __a, __a, __a, __a, __a}; diff --git a/clang/test/Headers/wasm.c b/clang/test/Headers/wasm.c index 30b15dd..f5076ae 100644 --- a/clang/test/Headers/wasm.c +++ b/clang/test/Headers/wasm.c @@ -386,6 +386,54 @@ v128_t test_f64x2_const() { return wasm_f64x2_const(0, 1); } +// CHECK-LABEL: @test_i8x16_const_splat( +// CHECK-NEXT: entry: +// CHECK-NEXT: ret <4 x i32> +// +v128_t test_i8x16_const_splat() { + return wasm_i8x16_const_splat(42); +} + +// CHECK-LABEL: @test_i16x8_const_splat( +// CHECK-NEXT: entry: +// CHECK-NEXT: ret <4 x i32> +// +v128_t test_i16x8_const_splat() { + return wasm_i16x8_const_splat(42); +} + +// CHECK-LABEL: @test_i32x4_const_splat( +// CHECK-NEXT: entry: +// CHECK-NEXT: ret <4 x i32> +// +v128_t test_i32x4_const_splat() { + return wasm_i32x4_const_splat(42); +} + +// CHECK-LABEL: @test_i64x2_const_splat( +// CHECK-NEXT: entry: +// CHECK-NEXT: ret <4 x i32> +// +v128_t test_i64x2_const_splat() { + return wasm_i64x2_const_splat(42); +} + +// CHECK-LABEL: @test_f32x4_const_splat( +// CHECK-NEXT: entry: +// CHECK-NEXT: ret <4 x i32> +// +v128_t test_f32x4_const_splat() { + return wasm_f32x4_const_splat(42); +} + +// CHECK-LABEL: @test_f64x2_const_splat( +// CHECK-NEXT: entry: +// CHECK-NEXT: ret <4 x i32> +// +v128_t test_f64x2_const_splat() { + return wasm_f64x2_const_splat(42); +} + // CHECK-LABEL: @test_i8x16_splat( // CHECK-NEXT: entry: // CHECK-NEXT: [[VECINIT_I:%.*]] = insertelement <16 x i8> undef, i8 [[A:%.*]], i32 0 -- 2.7.4