Index: clang/lib/Basic/Targets.cpp =================================================================== --- clang/lib/Basic/Targets.cpp +++ clang/lib/Basic/Targets.cpp @@ -6231,7 +6231,7 @@ V8_1A = 1; if (Feature == "+v8.2a") V8_2A = 1; - if (Feature == "+fp16") + if (Feature == "+fullfp16") HasFP16 = 1; } Index: clang/test/CodeGen/aarch64-v8.2a-neon-intrinsics.c =================================================================== --- clang/test/CodeGen/aarch64-v8.2a-neon-intrinsics.c +++ clang/test/CodeGen/aarch64-v8.2a-neon-intrinsics.c @@ -1,4 +1,4 @@ -// RUN: %clang_cc1 -triple arm64-none-linux-gnu -target-feature +neon -target-feature +fp16 -target-feature +v8.2a \ +// RUN: %clang_cc1 -triple arm64-none-linux-gnu -target-feature +neon -target-feature +fullfp16 -target-feature +v8.2a\ // RUN: -fallow-half-arguments-and-returns -S -emit-llvm -o - %s \ // RUN: | opt -S -mem2reg \ // RUN: | FileCheck %s Index: clang/test/CodeGen/arm_neon_intrinsics.c =================================================================== --- clang/test/CodeGen/arm_neon_intrinsics.c +++ clang/test/CodeGen/arm_neon_intrinsics.c @@ -3897,9 +3897,8 @@ // CHECK-LABEL: @test_vld1q_f16( // CHECK: [[TMP0:%.*]] = bitcast half* %a to i8* -// CHECK: [[VLD1:%.*]] = call <8 x i16> @llvm.arm.neon.vld1.v8i16.p0i8(i8* [[TMP0]], i32 2) -// CHECK: [[TMP1:%.*]] = bitcast <8 x i16> [[VLD1]] to <8 x half> -// CHECK: ret <8 x half> [[TMP1]] +// CHECK: [[VLD1:%.*]] = call <8 x half> @llvm.arm.neon.vld1.v8f16.p0i8(i8* [[TMP0]], i32 2) +// CHECK: ret <8 x half> [[VLD1]] float16x8_t test_vld1q_f16(float16_t const * a) { return vld1q_f16(a); } @@ -3991,9 +3990,8 @@ // CHECK-LABEL: @test_vld1_f16( // CHECK: [[TMP0:%.*]] = bitcast half* %a to i8* -// CHECK: [[VLD1:%.*]] = call <4 x i16> @llvm.arm.neon.vld1.v4i16.p0i8(i8* [[TMP0]], i32 2) -// CHECK: [[TMP1:%.*]] = bitcast <4 x i16> [[VLD1]] to <4 x half> -// CHECK: ret <4 x half> [[TMP1]] +// CHECK: [[VLD1:%.*]] = call <4 x half> @llvm.arm.neon.vld1.v4f16.p0i8(i8* [[TMP0]], i32 2) +// CHECK: ret <4 x half> [[VLD1]] float16x4_t test_vld1_f16(float16_t const * a) { return vld1_f16(a); } @@ -4107,12 +4105,11 @@ // CHECK-LABEL: @test_vld1q_dup_f16( // CHECK: [[TMP0:%.*]] = bitcast half* %a to i8* -// CHECK: [[TMP1:%.*]] = bitcast i8* [[TMP0]] to i16* -// CHECK: [[TMP2:%.*]] = load i16, i16* [[TMP1]], align 2 -// CHECK: [[TMP3:%.*]] = insertelement <8 x i16> undef, i16 [[TMP2]], i32 0 -// CHECK: [[LANE:%.*]] = shufflevector <8 x i16> [[TMP3]], <8 x i16> [[TMP3]], <8 x i32> zeroinitializer -// CHECK: [[TMP4:%.*]] = bitcast <8 x i16> [[LANE]] to <8 x half> -// CHECK: ret <8 x half> [[TMP4]] +// CHECK: [[TMP1:%.*]] = bitcast i8* [[TMP0]] to half* +// CHECK: [[TMP2:%.*]] = load half, half* [[TMP1]], align 2 +// CHECK: [[TMP3:%.*]] = insertelement <8 x half> undef, half [[TMP2]], i32 0 +// CHECK: [[LANE:%.*]] = shufflevector <8 x half> [[TMP3]], <8 x half> [[TMP3]], <8 x i32> zeroinitializer +// CHECK: ret <8 x half> [[LANE]] float16x8_t test_vld1q_dup_f16(float16_t const * a) { return vld1q_dup_f16(a); } @@ -4234,12 +4231,11 @@ // CHECK-LABEL: @test_vld1_dup_f16( // CHECK: [[TMP0:%.*]] = bitcast half* %a to i8* -// CHECK: [[TMP1:%.*]] = bitcast i8* [[TMP0]] to i16* -// CHECK: [[TMP2:%.*]] = load i16, i16* [[TMP1]], align 2 -// CHECK: [[TMP3:%.*]] = insertelement <4 x i16> undef, i16 [[TMP2]], i32 0 -// CHECK: [[LANE:%.*]] = shufflevector <4 x i16> [[TMP3]], <4 x i16> [[TMP3]], <4 x i32> zeroinitializer -// CHECK: [[TMP4:%.*]] = bitcast <4 x i16> [[LANE]] to <4 x half> -// CHECK: ret <4 x half> [[TMP4]] +// CHECK: [[TMP1:%.*]] = bitcast i8* [[TMP0]] to half* +// CHECK: [[TMP2:%.*]] = load half, half* [[TMP1]], align 2 +// CHECK: [[TMP3:%.*]] = insertelement <4 x half> undef, half [[TMP2]], i32 0 +// CHECK: [[LANE:%.*]] = shufflevector <4 x half> [[TMP3]], <4 x half> [[TMP3]], <4 x i32> zeroinitializer +// CHECK: ret <4 x half> [[LANE]] float16x4_t test_vld1_dup_f16(float16_t const * a) { return vld1_dup_f16(a); } @@ -4366,12 +4362,11 @@ // CHECK-LABEL: @test_vld1q_lane_f16( // CHECK: [[TMP0:%.*]] = bitcast half* %a to i8* // CHECK: [[TMP1:%.*]] = bitcast <8 x half> %b to <16 x i8> -// CHECK: [[TMP2:%.*]] = bitcast <16 x i8> [[TMP1]] to <8 x i16> -// CHECK: [[TMP3:%.*]] = bitcast i8* [[TMP0]] to i16* -// CHECK: [[TMP4:%.*]] = load i16, i16* [[TMP3]], align 2 -// CHECK: [[VLD1_LANE:%.*]] = insertelement <8 x i16> [[TMP2]], i16 [[TMP4]], i32 7 -// CHECK: [[TMP5:%.*]] = bitcast <8 x i16> [[VLD1_LANE]] to <8 x half> -// CHECK: ret <8 x half> [[TMP5]] +// CHECK: [[TMP2:%.*]] = bitcast <16 x i8> [[TMP1]] to <8 x half> +// CHECK: [[TMP3:%.*]] = bitcast i8* [[TMP0]] to half* +// CHECK: [[TMP4:%.*]] = load half, half* [[TMP3]], align 2 +// CHECK: [[VLD1_LANE:%.*]] = insertelement <8 x half> [[TMP2]], half [[TMP4]], i32 7 +// CHECK: ret <8 x half> [[VLD1_LANE]] float16x8_t test_vld1q_lane_f16(float16_t const * a, float16x8_t b) { return vld1q_lane_f16(a, b, 7); } @@ -4499,12 +4494,11 @@ // CHECK-LABEL: @test_vld1_lane_f16( // CHECK: [[TMP0:%.*]] = bitcast half* %a to i8* // CHECK: [[TMP1:%.*]] = bitcast <4 x half> %b to <8 x i8> -// CHECK: [[TMP2:%.*]] = bitcast <8 x i8> [[TMP1]] to <4 x i16> -// CHECK: [[TMP3:%.*]] = bitcast i8* [[TMP0]] to i16* -// CHECK: [[TMP4:%.*]] = load i16, i16* [[TMP3]], align 2 -// CHECK: [[VLD1_LANE:%.*]] = insertelement <4 x i16> [[TMP2]], i16 [[TMP4]], i32 3 -// CHECK: [[TMP5:%.*]] = bitcast <4 x i16> [[VLD1_LANE]] to <4 x half> -// CHECK: ret <4 x half> [[TMP5]] +// CHECK: [[TMP2:%.*]] = bitcast <8 x i8> [[TMP1]] to <4 x half> +// CHECK: [[TMP3:%.*]] = bitcast i8* [[TMP0]] to half* +// CHECK: [[TMP4:%.*]] = load half, half* [[TMP3]], align 2 +// CHECK: [[VLD1_LANE:%.*]] = insertelement <4 x half> [[TMP2]], half [[TMP4]], i32 3 +// CHECK: ret <4 x half> [[VLD1_LANE]] float16x4_t test_vld1_lane_f16(float16_t const * a, float16x4_t b) { return vld1_lane_f16(a, b, 3); } @@ -4597,7 +4591,7 @@ // CHECK: [[__RET:%.*]] = alloca %struct.float16x8x2_t, align 16 // CHECK: [[TMP0:%.*]] = bitcast %struct.float16x8x2_t* [[__RET]] to i8* // CHECK: [[TMP1:%.*]] = bitcast half* %a to i8* -// CHECK: [[VLD2Q_V:%.*]] = call { <8 x i16>, <8 x i16> +// CHECK: [[VLD2Q_V:%.*]] = call { <8 x half>, <8 x half> float16x8x2_t test_vld2q_f16(float16_t const * a) { return vld2q_f16(a); } @@ -4702,7 +4696,7 @@ // CHECK: [[__RET:%.*]] = alloca %struct.float16x4x2_t, align 8 // CHECK: [[TMP0:%.*]] = bitcast %struct.float16x4x2_t* [[__RET]] to i8* // CHECK: [[TMP1:%.*]] = bitcast half* %a to i8* -// CHECK: [[VLD2_V:%.*]] = call { <4 x i16>, <4 x i16> +// CHECK: [[VLD2_V:%.*]] = call { <4 x half>, <4 x half> float16x4x2_t test_vld2_f16(float16_t const * a) { return vld2_f16(a); } @@ -4807,7 +4801,7 @@ // CHECK: [[__RET:%.*]] = alloca %struct.float16x4x2_t, align 8 // CHECK: [[TMP0:%.*]] = bitcast %struct.float16x4x2_t* [[__RET]] to i8* // CHECK: [[TMP1:%.*]] = bitcast half* %a to i8* -// CHECK: [[VLD_DUP:%.*]] = call { <4 x i16>, <4 x i16> +// CHECK: [[VLD_DUP:%.*]] = call { <4 x half>, <4 x half> float16x4x2_t test_vld2_dup_f16(float16_t const * a) { return vld2_dup_f16(a); } @@ -4966,9 +4960,9 @@ // CHECK: [[ARRAYIDX2:%.*]] = getelementptr inbounds [2 x <8 x half>], [2 x <8 x half>]* [[VAL1]], i32 0, i32 1 // CHECK: [[TMP7:%.*]] = load <8 x half>, <8 x half>* [[ARRAYIDX2]], align 16 // CHECK: [[TMP8:%.*]] = bitcast <8 x half> [[TMP7]] to <16 x i8> -// CHECK: [[TMP9:%.*]] = bitcast <16 x i8> [[TMP6]] to <8 x i16> -// CHECK: [[TMP10:%.*]] = bitcast <16 x i8> [[TMP8]] to <8 x i16> -// CHECK: [[VLD2Q_LANE_V:%.*]] = call { <8 x i16>, <8 x i16> +// CHECK: [[TMP9:%.*]] = bitcast <16 x i8> [[TMP6]] to <8 x half> +// CHECK: [[TMP10:%.*]] = bitcast <16 x i8> [[TMP8]] to <8 x half> +// CHECK: [[VLD2Q_LANE_V:%.*]] = call { <8 x half>, <8 x half> float16x8x2_t test_vld2q_lane_f16(float16_t const * a, float16x8x2_t b) { return vld2q_lane_f16(a, b, 7); } @@ -5199,9 +5193,9 @@ // CHECK: [[ARRAYIDX2:%.*]] = getelementptr inbounds [2 x <4 x half>], [2 x <4 x half>]* [[VAL1]], i32 0, i32 1 // CHECK: [[TMP7:%.*]] = load <4 x half>, <4 x half>* [[ARRAYIDX2]], align 8 // CHECK: [[TMP8:%.*]] = bitcast <4 x half> [[TMP7]] to <8 x i8> -// CHECK: [[TMP9:%.*]] = bitcast <8 x i8> [[TMP6]] to <4 x i16> -// CHECK: [[TMP10:%.*]] = bitcast <8 x i8> [[TMP8]] to <4 x i16> -// CHECK: [[VLD2_LANE_V:%.*]] = call { <4 x i16>, <4 x i16> +// CHECK: [[TMP9:%.*]] = bitcast <8 x i8> [[TMP6]] to <4 x half> +// CHECK: [[TMP10:%.*]] = bitcast <8 x i8> [[TMP8]] to <4 x half> +// CHECK: [[VLD2_LANE_V:%.*]] = call { <4 x half>, <4 x half> float16x4x2_t test_vld2_lane_f16(float16_t const * a, float16x4x2_t b) { return vld2_lane_f16(a, b, 3); } @@ -5338,7 +5332,7 @@ // CHECK: [[__RET:%.*]] = alloca %struct.float16x8x3_t, align 16 // CHECK: [[TMP0:%.*]] = bitcast %struct.float16x8x3_t* [[__RET]] to i8* // CHECK: [[TMP1:%.*]] = bitcast half* %a to i8* -// CHECK: [[VLD3Q_V:%.*]] = call { <8 x i16>, <8 x i16>, <8 x i16> +// CHECK: [[VLD3Q_V:%.*]] = call { <8 x half>, <8 x half>, <8 x half> float16x8x3_t test_vld3q_f16(float16_t const * a) { return vld3q_f16(a); } @@ -5443,7 +5437,7 @@ // CHECK: [[__RET:%.*]] = alloca %struct.float16x4x3_t, align 8 // CHECK: [[TMP0:%.*]] = bitcast %struct.float16x4x3_t* [[__RET]] to i8* // CHECK: [[TMP1:%.*]] = bitcast half* %a to i8* -// CHECK: [[VLD3_V:%.*]] = call { <4 x i16>, <4 x i16>, <4 x i16> +// CHECK: [[VLD3_V:%.*]] = call { <4 x half>, <4 x half>, <4 x half> float16x4x3_t test_vld3_f16(float16_t const * a) { return vld3_f16(a); } @@ -5548,7 +5542,7 @@ // CHECK: [[__RET:%.*]] = alloca %struct.float16x4x3_t, align 8 // CHECK: [[TMP0:%.*]] = bitcast %struct.float16x4x3_t* [[__RET]] to i8* // CHECK: [[TMP1:%.*]] = bitcast half* %a to i8* -// CHECK: [[VLD_DUP:%.*]] = call { <4 x i16>, <4 x i16>, <4 x i16> +// CHECK: [[VLD_DUP:%.*]] = call { <4 x half>, <4 x half>, <4 x half> float16x4x3_t test_vld3_dup_f16(float16_t const * a) { return vld3_dup_f16(a); } @@ -5731,10 +5725,10 @@ // CHECK: [[ARRAYIDX4:%.*]] = getelementptr inbounds [3 x <8 x half>], [3 x <8 x half>]* [[VAL3]], i32 0, i32 2 // CHECK: [[TMP9:%.*]] = load <8 x half>, <8 x half>* [[ARRAYIDX4]], align 16 // CHECK: [[TMP10:%.*]] = bitcast <8 x half> [[TMP9]] to <16 x i8> -// CHECK: [[TMP11:%.*]] = bitcast <16 x i8> [[TMP6]] to <8 x i16> -// CHECK: [[TMP12:%.*]] = bitcast <16 x i8> [[TMP8]] to <8 x i16> -// CHECK: [[TMP13:%.*]] = bitcast <16 x i8> [[TMP10]] to <8 x i16> -// CHECK: [[VLD3Q_LANE_V:%.*]] = call { <8 x i16>, <8 x i16>, <8 x i16> +// CHECK: [[TMP11:%.*]] = bitcast <16 x i8> [[TMP6]] to <8 x half> +// CHECK: [[TMP12:%.*]] = bitcast <16 x i8> [[TMP8]] to <8 x half> +// CHECK: [[TMP13:%.*]] = bitcast <16 x i8> [[TMP10]] to <8 x half> +// CHECK: [[VLD3Q_LANE_V:%.*]] = call { <8 x half>, <8 x half>, <8 x half> float16x8x3_t test_vld3q_lane_f16(float16_t const * a, float16x8x3_t b) { return vld3q_lane_f16(a, b, 7); } @@ -6005,10 +5999,10 @@ // CHECK: [[ARRAYIDX4:%.*]] = getelementptr inbounds [3 x <4 x half>], [3 x <4 x half>]* [[VAL3]], i32 0, i32 2 // CHECK: [[TMP9:%.*]] = load <4 x half>, <4 x half>* [[ARRAYIDX4]], align 8 // CHECK: [[TMP10:%.*]] = bitcast <4 x half> [[TMP9]] to <8 x i8> -// CHECK: [[TMP11:%.*]] = bitcast <8 x i8> [[TMP6]] to <4 x i16> -// CHECK: [[TMP12:%.*]] = bitcast <8 x i8> [[TMP8]] to <4 x i16> -// CHECK: [[TMP13:%.*]] = bitcast <8 x i8> [[TMP10]] to <4 x i16> -// CHECK: [[VLD3_LANE_V:%.*]] = call { <4 x i16>, <4 x i16>, <4 x i16> +// CHECK: [[TMP11:%.*]] = bitcast <8 x i8> [[TMP6]] to <4 x half> +// CHECK: [[TMP12:%.*]] = bitcast <8 x i8> [[TMP8]] to <4 x half> +// CHECK: [[TMP13:%.*]] = bitcast <8 x i8> [[TMP10]] to <4 x half> +// CHECK: [[VLD3_LANE_V:%.*]] = call { <4 x half>, <4 x half>, <4 x half> float16x4x3_t test_vld3_lane_f16(float16_t const * a, float16x4x3_t b) { return vld3_lane_f16(a, b, 3); } @@ -6158,7 +6152,7 @@ // CHECK: [[__RET:%.*]] = alloca %struct.float16x8x4_t, align 16 // CHECK: [[TMP0:%.*]] = bitcast %struct.float16x8x4_t* [[__RET]] to i8* // CHECK: [[TMP1:%.*]] = bitcast half* %a to i8* -// CHECK: [[VLD4Q_V:%.*]] = call { <8 x i16>, <8 x i16>, <8 x i16>, <8 x i16> +// CHECK: [[VLD4Q_V:%.*]] = call { <8 x half>, <8 x half>, <8 x half>, <8 x half> float16x8x4_t test_vld4q_f16(float16_t const * a) { return vld4q_f16(a); } @@ -6263,7 +6257,7 @@ // CHECK: [[__RET:%.*]] = alloca %struct.float16x4x4_t, align 8 // CHECK: [[TMP0:%.*]] = bitcast %struct.float16x4x4_t* [[__RET]] to i8* // CHECK: [[TMP1:%.*]] = bitcast half* %a to i8* -// CHECK: [[VLD4_V:%.*]] = call { <4 x i16>, <4 x i16>, <4 x i16>, <4 x i16> +// CHECK: [[VLD4_V:%.*]] = call { <4 x half>, <4 x half>, <4 x half>, <4 x half> float16x4x4_t test_vld4_f16(float16_t const * a) { return vld4_f16(a); } @@ -6368,7 +6362,7 @@ // CHECK: [[__RET:%.*]] = alloca %struct.float16x4x4_t, align 8 // CHECK: [[TMP0:%.*]] = bitcast %struct.float16x4x4_t* [[__RET]] to i8* // CHECK: [[TMP1:%.*]] = bitcast half* %a to i8* -// CHECK: [[VLD_DUP:%.*]] = call { <4 x i16>, <4 x i16>, <4 x i16>, <4 x i16> +// CHECK: [[VLD_DUP:%.*]] = call { <4 x half>, <4 x half>, <4 x half>, <4 x half> float16x4x4_t test_vld4_dup_f16(float16_t const * a) { return vld4_dup_f16(a); } @@ -6575,11 +6569,11 @@ // CHECK: [[ARRAYIDX6:%.*]] = getelementptr inbounds [4 x <8 x half>], [4 x <8 x half>]* [[VAL5]], i32 0, i32 3 // CHECK: [[TMP11:%.*]] = load <8 x half>, <8 x half>* [[ARRAYIDX6]], align 16 // CHECK: [[TMP12:%.*]] = bitcast <8 x half> [[TMP11]] to <16 x i8> -// CHECK: [[TMP13:%.*]] = bitcast <16 x i8> [[TMP6]] to <8 x i16> -// CHECK: [[TMP14:%.*]] = bitcast <16 x i8> [[TMP8]] to <8 x i16> -// CHECK: [[TMP15:%.*]] = bitcast <16 x i8> [[TMP10]] to <8 x i16> -// CHECK: [[TMP16:%.*]] = bitcast <16 x i8> [[TMP12]] to <8 x i16> -// CHECK: [[VLD4Q_LANE_V:%.*]] = call { <8 x i16>, <8 x i16>, <8 x i16>, <8 x i16> +// CHECK: [[TMP13:%.*]] = bitcast <16 x i8> [[TMP6]] to <8 x half> +// CHECK: [[TMP14:%.*]] = bitcast <16 x i8> [[TMP8]] to <8 x half> +// CHECK: [[TMP15:%.*]] = bitcast <16 x i8> [[TMP10]] to <8 x half> +// CHECK: [[TMP16:%.*]] = bitcast <16 x i8> [[TMP12]] to <8 x half> +// CHECK: [[VLD4Q_LANE_V:%.*]] = call { <8 x half>, <8 x half>, <8 x half>, <8 x half> float16x8x4_t test_vld4q_lane_f16(float16_t const * a, float16x8x4_t b) { return vld4q_lane_f16(a, b, 7); } @@ -6890,11 +6884,11 @@ // CHECK: [[ARRAYIDX6:%.*]] = getelementptr inbounds [4 x <4 x half>], [4 x <4 x half>]* [[VAL5]], i32 0, i32 3 // CHECK: [[TMP11:%.*]] = load <4 x half>, <4 x half>* [[ARRAYIDX6]], align 8 // CHECK: [[TMP12:%.*]] = bitcast <4 x half> [[TMP11]] to <8 x i8> -// CHECK: [[TMP13:%.*]] = bitcast <8 x i8> [[TMP6]] to <4 x i16> -// CHECK: [[TMP14:%.*]] = bitcast <8 x i8> [[TMP8]] to <4 x i16> -// CHECK: [[TMP15:%.*]] = bitcast <8 x i8> [[TMP10]] to <4 x i16> -// CHECK: [[TMP16:%.*]] = bitcast <8 x i8> [[TMP12]] to <4 x i16> -// CHECK: [[VLD4_LANE_V:%.*]] = call { <4 x i16>, <4 x i16>, <4 x i16>, <4 x i16> +// CHECK: [[TMP13:%.*]] = bitcast <8 x i8> [[TMP6]] to <4 x half> +// CHECK: [[TMP14:%.*]] = bitcast <8 x i8> [[TMP8]] to <4 x half> +// CHECK: [[TMP15:%.*]] = bitcast <8 x i8> [[TMP10]] to <4 x half> +// CHECK: [[TMP16:%.*]] = bitcast <8 x i8> [[TMP12]] to <4 x half> +// CHECK: [[VLD4_LANE_V:%.*]] = call { <4 x half>, <4 x half>, <4 x half>, <4 x half> float16x4x4_t test_vld4_lane_f16(float16_t const * a, float16x4x4_t b) { return vld4_lane_f16(a, b, 3); } @@ -15785,8 +15779,8 @@ // CHECK-LABEL: @test_vst1q_f16( // CHECK: [[TMP0:%.*]] = bitcast half* %a to i8* // CHECK: [[TMP1:%.*]] = bitcast <8 x half> %b to <16 x i8> -// CHECK: [[TMP2:%.*]] = bitcast <16 x i8> [[TMP1]] to <8 x i16> -// CHECK: call void @llvm.arm.neon.vst1.p0i8.v8i16(i8* [[TMP0]], <8 x i16> [[TMP2]], i32 2) +// CHECK: [[TMP2:%.*]] = bitcast <16 x i8> [[TMP1]] to <8 x half> +// CHECK: call void @llvm.arm.neon.vst1.p0i8.v8f16(i8* [[TMP0]], <8 x half> [[TMP2]], i32 2) // CHECK: ret void void test_vst1q_f16(float16_t * a, float16x8_t b) { vst1q_f16(a, b); @@ -15896,8 +15890,8 @@ // CHECK-LABEL: @test_vst1_f16( // CHECK: [[TMP0:%.*]] = bitcast half* %a to i8* // CHECK: [[TMP1:%.*]] = bitcast <4 x half> %b to <8 x i8> -// CHECK: [[TMP2:%.*]] = bitcast <8 x i8> [[TMP1]] to <4 x i16> -// CHECK: call void @llvm.arm.neon.vst1.p0i8.v4i16(i8* [[TMP0]], <4 x i16> [[TMP2]], i32 2) +// CHECK: [[TMP2:%.*]] = bitcast <8 x i8> [[TMP1]] to <4 x half> +// CHECK: call void @llvm.arm.neon.vst1.p0i8.v4f16(i8* [[TMP0]], <4 x half> [[TMP2]], i32 2) // CHECK: ret void void test_vst1_f16(float16_t * a, float16x4_t b) { vst1_f16(a, b); @@ -16019,10 +16013,10 @@ // CHECK-LABEL: @test_vst1q_lane_f16( // CHECK: [[TMP0:%.*]] = bitcast half* %a to i8* // CHECK: [[TMP1:%.*]] = bitcast <8 x half> %b to <16 x i8> -// CHECK: [[TMP2:%.*]] = bitcast <16 x i8> [[TMP1]] to <8 x i16> -// CHECK: [[TMP3:%.*]] = extractelement <8 x i16> [[TMP2]], i32 7 -// CHECK: [[TMP4:%.*]] = bitcast i8* [[TMP0]] to i16* -// CHECK: store i16 [[TMP3]], i16* [[TMP4]], align 2 +// CHECK: [[TMP2:%.*]] = bitcast <16 x i8> [[TMP1]] to <8 x half> +// CHECK: [[TMP3:%.*]] = extractelement <8 x half> [[TMP2]], i32 7 +// CHECK: [[TMP4:%.*]] = bitcast i8* [[TMP0]] to half* +// CHECK: store half [[TMP3]], half* [[TMP4]], align 2 // CHECK: ret void void test_vst1q_lane_f16(float16_t * a, float16x8_t b) { vst1q_lane_f16(a, b, 7); @@ -16151,10 +16145,10 @@ // CHECK-LABEL: @test_vst1_lane_f16( // CHECK: [[TMP0:%.*]] = bitcast half* %a to i8* // CHECK: [[TMP1:%.*]] = bitcast <4 x half> %b to <8 x i8> -// CHECK: [[TMP2:%.*]] = bitcast <8 x i8> [[TMP1]] to <4 x i16> -// CHECK: [[TMP3:%.*]] = extractelement <4 x i16> [[TMP2]], i32 3 -// CHECK: [[TMP4:%.*]] = bitcast i8* [[TMP0]] to i16* -// CHECK: store i16 [[TMP3]], i16* [[TMP4]], align 2 +// CHECK: [[TMP2:%.*]] = bitcast <8 x i8> [[TMP1]] to <4 x half> +// CHECK: [[TMP3:%.*]] = extractelement <4 x half> [[TMP2]], i32 3 +// CHECK: [[TMP4:%.*]] = bitcast i8* [[TMP0]] to half* +// CHECK: store half [[TMP3]], half* [[TMP4]], align 2 // CHECK: ret void void test_vst1_lane_f16(float16_t * a, float16x4_t b) { vst1_lane_f16(a, b, 3); @@ -16356,9 +16350,9 @@ // CHECK: [[ARRAYIDX2:%.*]] = getelementptr inbounds [2 x <8 x half>], [2 x <8 x half>]* [[VAL1]], i32 0, i32 1 // CHECK: [[TMP6:%.*]] = load <8 x half>, <8 x half>* [[ARRAYIDX2]], align 16 // CHECK: [[TMP7:%.*]] = bitcast <8 x half> [[TMP6]] to <16 x i8> -// CHECK: [[TMP8:%.*]] = bitcast <16 x i8> [[TMP5]] to <8 x i16> -// CHECK: [[TMP9:%.*]] = bitcast <16 x i8> [[TMP7]] to <8 x i16> -// CHECK: call void @llvm.arm.neon.vst2.p0i8.v8i16(i8* [[TMP3]], <8 x i16> [[TMP8]], <8 x i16> [[TMP9]], i32 2) +// CHECK: [[TMP8:%.*]] = bitcast <16 x i8> [[TMP5]] to <8 x half> +// CHECK: [[TMP9:%.*]] = bitcast <16 x i8> [[TMP7]] to <8 x half> +// CHECK: call void @llvm.arm.neon.vst2.p0i8.v8f16(i8* [[TMP3]], <8 x half> [[TMP8]], <8 x half> [[TMP9]], i32 2) // CHECK: ret void void test_vst2q_f16(float16_t * a, float16x8x2_t b) { vst2q_f16(a, b); @@ -16653,9 +16647,9 @@ // CHECK: [[ARRAYIDX2:%.*]] = getelementptr inbounds [2 x <4 x half>], [2 x <4 x half>]* [[VAL1]], i32 0, i32 1 // CHECK: [[TMP6:%.*]] = load <4 x half>, <4 x half>* [[ARRAYIDX2]], align 8 // CHECK: [[TMP7:%.*]] = bitcast <4 x half> [[TMP6]] to <8 x i8> -// CHECK: [[TMP8:%.*]] = bitcast <8 x i8> [[TMP5]] to <4 x i16> -// CHECK: [[TMP9:%.*]] = bitcast <8 x i8> [[TMP7]] to <4 x i16> -// CHECK: call void @llvm.arm.neon.vst2.p0i8.v4i16(i8* [[TMP3]], <4 x i16> [[TMP8]], <4 x i16> [[TMP9]], i32 2) +// CHECK: [[TMP8:%.*]] = bitcast <8 x i8> [[TMP5]] to <4 x half> +// CHECK: [[TMP9:%.*]] = bitcast <8 x i8> [[TMP7]] to <4 x half> +// CHECK: call void @llvm.arm.neon.vst2.p0i8.v4f16(i8* [[TMP3]], <4 x half> [[TMP8]], <4 x half> [[TMP9]], i32 2) // CHECK: ret void void test_vst2_f16(float16_t * a, float16x4x2_t b) { vst2_f16(a, b); @@ -16856,9 +16850,9 @@ // CHECK: [[ARRAYIDX2:%.*]] = getelementptr inbounds [2 x <8 x half>], [2 x <8 x half>]* [[VAL1]], i32 0, i32 1 // CHECK: [[TMP6:%.*]] = load <8 x half>, <8 x half>* [[ARRAYIDX2]], align 16 // CHECK: [[TMP7:%.*]] = bitcast <8 x half> [[TMP6]] to <16 x i8> -// CHECK: [[TMP8:%.*]] = bitcast <16 x i8> [[TMP5]] to <8 x i16> -// CHECK: [[TMP9:%.*]] = bitcast <16 x i8> [[TMP7]] to <8 x i16> -// CHECK: call void @llvm.arm.neon.vst2lane.p0i8.v8i16(i8* [[TMP3]], <8 x i16> [[TMP8]], <8 x i16> [[TMP9]], i32 7, i32 2) +// CHECK: [[TMP8:%.*]] = bitcast <16 x i8> [[TMP5]] to <8 x half> +// CHECK: [[TMP9:%.*]] = bitcast <16 x i8> [[TMP7]] to <8 x half> +// CHECK: call void @llvm.arm.neon.vst2lane.p0i8.v8f16(i8* [[TMP3]], <8 x half> [[TMP8]], <8 x half> [[TMP9]], i32 7, i32 2) // CHECK: ret void void test_vst2q_lane_f16(float16_t * a, float16x8x2_t b) { vst2q_lane_f16(a, b, 7); @@ -17080,9 +17074,9 @@ // CHECK: [[ARRAYIDX2:%.*]] = getelementptr inbounds [2 x <4 x half>], [2 x <4 x half>]* [[VAL1]], i32 0, i32 1 // CHECK: [[TMP6:%.*]] = load <4 x half>, <4 x half>* [[ARRAYIDX2]], align 8 // CHECK: [[TMP7:%.*]] = bitcast <4 x half> [[TMP6]] to <8 x i8> -// CHECK: [[TMP8:%.*]] = bitcast <8 x i8> [[TMP5]] to <4 x i16> -// CHECK: [[TMP9:%.*]] = bitcast <8 x i8> [[TMP7]] to <4 x i16> -// CHECK: call void @llvm.arm.neon.vst2lane.p0i8.v4i16(i8* [[TMP3]], <4 x i16> [[TMP8]], <4 x i16> [[TMP9]], i32 3, i32 2) +// CHECK: [[TMP8:%.*]] = bitcast <8 x i8> [[TMP5]] to <4 x half> +// CHECK: [[TMP9:%.*]] = bitcast <8 x i8> [[TMP7]] to <4 x half> +// CHECK: call void @llvm.arm.neon.vst2lane.p0i8.v4f16(i8* [[TMP3]], <4 x half> [[TMP8]], <4 x half> [[TMP9]], i32 3, i32 2) // CHECK: ret void void test_vst2_lane_f16(float16_t * a, float16x4x2_t b) { vst2_lane_f16(a, b, 3); @@ -17355,10 +17349,10 @@ // CHECK: [[ARRAYIDX4:%.*]] = getelementptr inbounds [3 x <8 x half>], [3 x <8 x half>]* [[VAL3]], i32 0, i32 2 // CHECK: [[TMP8:%.*]] = load <8 x half>, <8 x half>* [[ARRAYIDX4]], align 16 // CHECK: [[TMP9:%.*]] = bitcast <8 x half> [[TMP8]] to <16 x i8> -// CHECK: [[TMP10:%.*]] = bitcast <16 x i8> [[TMP5]] to <8 x i16> -// CHECK: [[TMP11:%.*]] = bitcast <16 x i8> [[TMP7]] to <8 x i16> -// CHECK: [[TMP12:%.*]] = bitcast <16 x i8> [[TMP9]] to <8 x i16> -// CHECK: call void @llvm.arm.neon.vst3.p0i8.v8i16(i8* [[TMP3]], <8 x i16> [[TMP10]], <8 x i16> [[TMP11]], <8 x i16> [[TMP12]], i32 2) +// CHECK: [[TMP10:%.*]] = bitcast <16 x i8> [[TMP5]] to <8 x half> +// CHECK: [[TMP11:%.*]] = bitcast <16 x i8> [[TMP7]] to <8 x half> +// CHECK: [[TMP12:%.*]] = bitcast <16 x i8> [[TMP9]] to <8 x half> +// CHECK: call void @llvm.arm.neon.vst3.p0i8.v8f16(i8* [[TMP3]], <8 x half> [[TMP10]], <8 x half> [[TMP11]], <8 x half> [[TMP12]], i32 2) // CHECK: ret void void test_vst3q_f16(float16_t * a, float16x8x3_t b) { vst3q_f16(a, b); @@ -17706,10 +17700,10 @@ // CHECK: [[ARRAYIDX4:%.*]] = getelementptr inbounds [3 x <4 x half>], [3 x <4 x half>]* [[VAL3]], i32 0, i32 2 // CHECK: [[TMP8:%.*]] = load <4 x half>, <4 x half>* [[ARRAYIDX4]], align 8 // CHECK: [[TMP9:%.*]] = bitcast <4 x half> [[TMP8]] to <8 x i8> -// CHECK: [[TMP10:%.*]] = bitcast <8 x i8> [[TMP5]] to <4 x i16> -// CHECK: [[TMP11:%.*]] = bitcast <8 x i8> [[TMP7]] to <4 x i16> -// CHECK: [[TMP12:%.*]] = bitcast <8 x i8> [[TMP9]] to <4 x i16> -// CHECK: call void @llvm.arm.neon.vst3.p0i8.v4i16(i8* [[TMP3]], <4 x i16> [[TMP10]], <4 x i16> [[TMP11]], <4 x i16> [[TMP12]], i32 2) +// CHECK: [[TMP10:%.*]] = bitcast <8 x i8> [[TMP5]] to <4 x half> +// CHECK: [[TMP11:%.*]] = bitcast <8 x i8> [[TMP7]] to <4 x half> +// CHECK: [[TMP12:%.*]] = bitcast <8 x i8> [[TMP9]] to <4 x half> +// CHECK: call void @llvm.arm.neon.vst3.p0i8.v4f16(i8* [[TMP3]], <4 x half> [[TMP10]], <4 x half> [[TMP11]], <4 x half> [[TMP12]], i32 2) // CHECK: ret void void test_vst3_f16(float16_t * a, float16x4x3_t b) { vst3_f16(a, b); @@ -17947,10 +17941,10 @@ // CHECK: [[ARRAYIDX4:%.*]] = getelementptr inbounds [3 x <8 x half>], [3 x <8 x half>]* [[VAL3]], i32 0, i32 2 // CHECK: [[TMP8:%.*]] = load <8 x half>, <8 x half>* [[ARRAYIDX4]], align 16 // CHECK: [[TMP9:%.*]] = bitcast <8 x half> [[TMP8]] to <16 x i8> -// CHECK: [[TMP10:%.*]] = bitcast <16 x i8> [[TMP5]] to <8 x i16> -// CHECK: [[TMP11:%.*]] = bitcast <16 x i8> [[TMP7]] to <8 x i16> -// CHECK: [[TMP12:%.*]] = bitcast <16 x i8> [[TMP9]] to <8 x i16> -// CHECK: call void @llvm.arm.neon.vst3lane.p0i8.v8i16(i8* [[TMP3]], <8 x i16> [[TMP10]], <8 x i16> [[TMP11]], <8 x i16> [[TMP12]], i32 7, i32 2) +// CHECK: [[TMP10:%.*]] = bitcast <16 x i8> [[TMP5]] to <8 x half> +// CHECK: [[TMP11:%.*]] = bitcast <16 x i8> [[TMP7]] to <8 x half> +// CHECK: [[TMP12:%.*]] = bitcast <16 x i8> [[TMP9]] to <8 x half> +// CHECK: call void @llvm.arm.neon.vst3lane.p0i8.v8f16(i8* [[TMP3]], <8 x half> [[TMP10]], <8 x half> [[TMP11]], <8 x half> [[TMP12]], i32 7, i32 2) // CHECK: ret void void test_vst3q_lane_f16(float16_t * a, float16x8x3_t b) { vst3q_lane_f16(a, b, 7); @@ -18212,10 +18206,10 @@ // CHECK: [[ARRAYIDX4:%.*]] = getelementptr inbounds [3 x <4 x half>], [3 x <4 x half>]* [[VAL3]], i32 0, i32 2 // CHECK: [[TMP8:%.*]] = load <4 x half>, <4 x half>* [[ARRAYIDX4]], align 8 // CHECK: [[TMP9:%.*]] = bitcast <4 x half> [[TMP8]] to <8 x i8> -// CHECK: [[TMP10:%.*]] = bitcast <8 x i8> [[TMP5]] to <4 x i16> -// CHECK: [[TMP11:%.*]] = bitcast <8 x i8> [[TMP7]] to <4 x i16> -// CHECK: [[TMP12:%.*]] = bitcast <8 x i8> [[TMP9]] to <4 x i16> -// CHECK: call void @llvm.arm.neon.vst3lane.p0i8.v4i16(i8* [[TMP3]], <4 x i16> [[TMP10]], <4 x i16> [[TMP11]], <4 x i16> [[TMP12]], i32 3, i32 2) +// CHECK: [[TMP10:%.*]] = bitcast <8 x i8> [[TMP5]] to <4 x half> +// CHECK: [[TMP11:%.*]] = bitcast <8 x i8> [[TMP7]] to <4 x half> +// CHECK: [[TMP12:%.*]] = bitcast <8 x i8> [[TMP9]] to <4 x half> +// CHECK: call void @llvm.arm.neon.vst3lane.p0i8.v4f16(i8* [[TMP3]], <4 x half> [[TMP10]], <4 x half> [[TMP11]], <4 x half> [[TMP12]], i32 3, i32 2) // CHECK: ret void void test_vst3_lane_f16(float16_t * a, float16x4x3_t b) { vst3_lane_f16(a, b, 3); @@ -18531,11 +18525,11 @@ // CHECK: [[ARRAYIDX6:%.*]] = getelementptr inbounds [4 x <8 x half>], [4 x <8 x half>]* [[VAL5]], i32 0, i32 3 // CHECK: [[TMP10:%.*]] = load <8 x half>, <8 x half>* [[ARRAYIDX6]], align 16 // CHECK: [[TMP11:%.*]] = bitcast <8 x half> [[TMP10]] to <16 x i8> -// CHECK: [[TMP12:%.*]] = bitcast <16 x i8> [[TMP5]] to <8 x i16> -// CHECK: [[TMP13:%.*]] = bitcast <16 x i8> [[TMP7]] to <8 x i16> -// CHECK: [[TMP14:%.*]] = bitcast <16 x i8> [[TMP9]] to <8 x i16> -// CHECK: [[TMP15:%.*]] = bitcast <16 x i8> [[TMP11]] to <8 x i16> -// CHECK: call void @llvm.arm.neon.vst4.p0i8.v8i16(i8* [[TMP3]], <8 x i16> [[TMP12]], <8 x i16> [[TMP13]], <8 x i16> [[TMP14]], <8 x i16> [[TMP15]], i32 2) +// CHECK: [[TMP12:%.*]] = bitcast <16 x i8> [[TMP5]] to <8 x half> +// CHECK: [[TMP13:%.*]] = bitcast <16 x i8> [[TMP7]] to <8 x half> +// CHECK: [[TMP14:%.*]] = bitcast <16 x i8> [[TMP9]] to <8 x half> +// CHECK: [[TMP15:%.*]] = bitcast <16 x i8> [[TMP11]] to <8 x half> +// CHECK: call void @llvm.arm.neon.vst4.p0i8.v8f16(i8* [[TMP3]], <8 x half> [[TMP12]], <8 x half> [[TMP13]], <8 x half> [[TMP14]], <8 x half> [[TMP15]], i32 2) // CHECK: ret void void test_vst4q_f16(float16_t * a, float16x8x4_t b) { vst4q_f16(a, b); @@ -18936,11 +18930,11 @@ // CHECK: [[ARRAYIDX6:%.*]] = getelementptr inbounds [4 x <4 x half>], [4 x <4 x half>]* [[VAL5]], i32 0, i32 3 // CHECK: [[TMP10:%.*]] = load <4 x half>, <4 x half>* [[ARRAYIDX6]], align 8 // CHECK: [[TMP11:%.*]] = bitcast <4 x half> [[TMP10]] to <8 x i8> -// CHECK: [[TMP12:%.*]] = bitcast <8 x i8> [[TMP5]] to <4 x i16> -// CHECK: [[TMP13:%.*]] = bitcast <8 x i8> [[TMP7]] to <4 x i16> -// CHECK: [[TMP14:%.*]] = bitcast <8 x i8> [[TMP9]] to <4 x i16> -// CHECK: [[TMP15:%.*]] = bitcast <8 x i8> [[TMP11]] to <4 x i16> -// CHECK: call void @llvm.arm.neon.vst4.p0i8.v4i16(i8* [[TMP3]], <4 x i16> [[TMP12]], <4 x i16> [[TMP13]], <4 x i16> [[TMP14]], <4 x i16> [[TMP15]], i32 2) +// CHECK: [[TMP12:%.*]] = bitcast <8 x i8> [[TMP5]] to <4 x half> +// CHECK: [[TMP13:%.*]] = bitcast <8 x i8> [[TMP7]] to <4 x half> +// CHECK: [[TMP14:%.*]] = bitcast <8 x i8> [[TMP9]] to <4 x half> +// CHECK: [[TMP15:%.*]] = bitcast <8 x i8> [[TMP11]] to <4 x half> +// CHECK: call void @llvm.arm.neon.vst4.p0i8.v4f16(i8* [[TMP3]], <4 x half> [[TMP12]], <4 x half> [[TMP13]], <4 x half> [[TMP14]], <4 x half> [[TMP15]], i32 2) // CHECK: ret void void test_vst4_f16(float16_t * a, float16x4x4_t b) { vst4_f16(a, b); @@ -19215,11 +19209,11 @@ // CHECK: [[ARRAYIDX6:%.*]] = getelementptr inbounds [4 x <8 x half>], [4 x <8 x half>]* [[VAL5]], i32 0, i32 3 // CHECK: [[TMP10:%.*]] = load <8 x half>, <8 x half>* [[ARRAYIDX6]], align 16 // CHECK: [[TMP11:%.*]] = bitcast <8 x half> [[TMP10]] to <16 x i8> -// CHECK: [[TMP12:%.*]] = bitcast <16 x i8> [[TMP5]] to <8 x i16> -// CHECK: [[TMP13:%.*]] = bitcast <16 x i8> [[TMP7]] to <8 x i16> -// CHECK: [[TMP14:%.*]] = bitcast <16 x i8> [[TMP9]] to <8 x i16> -// CHECK: [[TMP15:%.*]] = bitcast <16 x i8> [[TMP11]] to <8 x i16> -// CHECK: call void @llvm.arm.neon.vst4lane.p0i8.v8i16(i8* [[TMP3]], <8 x i16> [[TMP12]], <8 x i16> [[TMP13]], <8 x i16> [[TMP14]], <8 x i16> [[TMP15]], i32 7, i32 2) +// CHECK: [[TMP12:%.*]] = bitcast <16 x i8> [[TMP5]] to <8 x half> +// CHECK: [[TMP13:%.*]] = bitcast <16 x i8> [[TMP7]] to <8 x half> +// CHECK: [[TMP14:%.*]] = bitcast <16 x i8> [[TMP9]] to <8 x half> +// CHECK: [[TMP15:%.*]] = bitcast <16 x i8> [[TMP11]] to <8 x half> +// CHECK: call void @llvm.arm.neon.vst4lane.p0i8.v8f16(i8* [[TMP3]], <8 x half> [[TMP12]], <8 x half> [[TMP13]], <8 x half> [[TMP14]], <8 x half> [[TMP15]], i32 7, i32 2) // CHECK: ret void void test_vst4q_lane_f16(float16_t * a, float16x8x4_t b) { vst4q_lane_f16(a, b, 7); @@ -19521,11 +19515,11 @@ // CHECK: [[ARRAYIDX6:%.*]] = getelementptr inbounds [4 x <4 x half>], [4 x <4 x half>]* [[VAL5]], i32 0, i32 3 // CHECK: [[TMP10:%.*]] = load <4 x half>, <4 x half>* [[ARRAYIDX6]], align 8 // CHECK: [[TMP11:%.*]] = bitcast <4 x half> [[TMP10]] to <8 x i8> -// CHECK: [[TMP12:%.*]] = bitcast <8 x i8> [[TMP5]] to <4 x i16> -// CHECK: [[TMP13:%.*]] = bitcast <8 x i8> [[TMP7]] to <4 x i16> -// CHECK: [[TMP14:%.*]] = bitcast <8 x i8> [[TMP9]] to <4 x i16> -// CHECK: [[TMP15:%.*]] = bitcast <8 x i8> [[TMP11]] to <4 x i16> -// CHECK: call void @llvm.arm.neon.vst4lane.p0i8.v4i16(i8* [[TMP3]], <4 x i16> [[TMP12]], <4 x i16> [[TMP13]], <4 x i16> [[TMP14]], <4 x i16> [[TMP15]], i32 3, i32 2) +// CHECK: [[TMP12:%.*]] = bitcast <8 x i8> [[TMP5]] to <4 x half> +// CHECK: [[TMP13:%.*]] = bitcast <8 x i8> [[TMP7]] to <4 x half> +// CHECK: [[TMP14:%.*]] = bitcast <8 x i8> [[TMP9]] to <4 x half> +// CHECK: [[TMP15:%.*]] = bitcast <8 x i8> [[TMP11]] to <4 x half> +// CHECK: call void @llvm.arm.neon.vst4lane.p0i8.v4f16(i8* [[TMP3]], <4 x half> [[TMP12]], <4 x half> [[TMP13]], <4 x half> [[TMP14]], <4 x half> [[TMP15]], i32 3, i32 2) // CHECK: ret void void test_vst4_lane_f16(float16_t * a, float16x4x4_t b) { vst4_lane_f16(a, b, 3);