maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/clang/test/CodeGen/arm-mve-intrinsics/ternary.c

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

642 lines
28 KiB
C
Raw Normal View History

[ARM,MVE] Add intrinsics and isel for MVE fused multiply-add. Summary: This adds the ACLE intrinsic family for the VFMA and VFMS instructions, which perform fused multiply-add on vectors of floats. I've represented the unpredicated versions in IR using the cross- platform `@llvm.fma` IR intrinsic. We already had isel rules to convert one of those into a vector VFMA in the simplest possible way; but we didn't have rules to detect a negated argument and turn it into VFMS, or rules to detect a splat argument and turn it into one of the two vector/scalar forms of the instruction. Now we have all of those. The predicated form uses a target-specific intrinsic as usual, but I've stuck to just one, for a predicated FMA. The subtraction and splat versions are code-generated by passing an fneg or a splat as one of its operands, the same way as the unpredicated version. In arm_mve_defs.h, I've had to introduce a tiny extra piece of infrastructure: a record `id` for use in codegen dags which implements the identity function. (Just because you can't declare a Tablegen value of type dag which is //only// a `$varname`: you have to wrap it in something. Now I can write `(id $varname)` to get the same effect.) Reviewers: dmgreen, MarkMurrayARM, miyuki, ostannard Reviewed By: dmgreen Subscribers: kristof.beyls, hiraditya, danielkiss, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D75998
2020-03-12 17:57:48 +08:00
// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
// RUN: %clang_cc1 -triple thumbv8.1m.main-arm-none-eabi -target-feature +mve.fp -mfloat-abi hard -fallow-half-arguments-and-returns -O0 -disable-O0-optnone -S -emit-llvm -o - %s | opt -S -sroa | FileCheck %s
// RUN: %clang_cc1 -triple thumbv8.1m.main-arm-none-eabi -target-feature +mve.fp -mfloat-abi hard -fallow-half-arguments-and-returns -O0 -disable-O0-optnone -DPOLYMORPHIC -S -emit-llvm -o - %s | opt -S -sroa | FileCheck %s
#include <arm_mve.h>
// CHECK-LABEL: @test_vfmaq_f16(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = call <8 x half> @llvm.fma.v8f16(<8 x half> [[B:%.*]], <8 x half> [[C:%.*]], <8 x half> [[A:%.*]])
// CHECK-NEXT: ret <8 x half> [[TMP0]]
//
float16x8_t test_vfmaq_f16(float16x8_t a, float16x8_t b, float16x8_t c) {
#ifdef POLYMORPHIC
return vfmaq(a, b, c);
#else /* POLYMORPHIC */
return vfmaq_f16(a, b, c);
#endif /* POLYMORPHIC */
}
// CHECK-LABEL: @test_vfmaq_f32(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = call <4 x float> @llvm.fma.v4f32(<4 x float> [[B:%.*]], <4 x float> [[C:%.*]], <4 x float> [[A:%.*]])
// CHECK-NEXT: ret <4 x float> [[TMP0]]
//
float32x4_t test_vfmaq_f32(float32x4_t a, float32x4_t b, float32x4_t c) {
#ifdef POLYMORPHIC
return vfmaq(a, b, c);
#else /* POLYMORPHIC */
return vfmaq_f32(a, b, c);
#endif /* POLYMORPHIC */
}
// CHECK-LABEL: @test_vfmaq_n_f16(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = bitcast float [[C_COERCE:%.*]] to i32
// CHECK-NEXT: [[TMP_0_EXTRACT_TRUNC:%.*]] = trunc i32 [[TMP0]] to i16
// CHECK-NEXT: [[TMP1:%.*]] = bitcast i16 [[TMP_0_EXTRACT_TRUNC]] to half
// CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <8 x half> undef, half [[TMP1]], i32 0
// CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <8 x half> [[DOTSPLATINSERT]], <8 x half> undef, <8 x i32> zeroinitializer
// CHECK-NEXT: [[TMP2:%.*]] = call <8 x half> @llvm.fma.v8f16(<8 x half> [[B:%.*]], <8 x half> [[DOTSPLAT]], <8 x half> [[A:%.*]])
// CHECK-NEXT: ret <8 x half> [[TMP2]]
//
float16x8_t test_vfmaq_n_f16(float16x8_t a, float16x8_t b, float16_t c) {
#ifdef POLYMORPHIC
return vfmaq(a, b, c);
#else /* POLYMORPHIC */
return vfmaq_n_f16(a, b, c);
#endif /* POLYMORPHIC */
}
// CHECK-LABEL: @test_vfmaq_n_f32(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <4 x float> undef, float [[C:%.*]], i32 0
// CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <4 x float> [[DOTSPLATINSERT]], <4 x float> undef, <4 x i32> zeroinitializer
// CHECK-NEXT: [[TMP0:%.*]] = call <4 x float> @llvm.fma.v4f32(<4 x float> [[B:%.*]], <4 x float> [[DOTSPLAT]], <4 x float> [[A:%.*]])
// CHECK-NEXT: ret <4 x float> [[TMP0]]
//
float32x4_t test_vfmaq_n_f32(float32x4_t a, float32x4_t b, float32_t c) {
#ifdef POLYMORPHIC
return vfmaq(a, b, c);
#else /* POLYMORPHIC */
return vfmaq_n_f32(a, b, c);
#endif /* POLYMORPHIC */
}
// CHECK-LABEL: @test_vfmasq_n_f16(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = bitcast float [[C_COERCE:%.*]] to i32
// CHECK-NEXT: [[TMP_0_EXTRACT_TRUNC:%.*]] = trunc i32 [[TMP0]] to i16
// CHECK-NEXT: [[TMP1:%.*]] = bitcast i16 [[TMP_0_EXTRACT_TRUNC]] to half
// CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <8 x half> undef, half [[TMP1]], i32 0
// CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <8 x half> [[DOTSPLATINSERT]], <8 x half> undef, <8 x i32> zeroinitializer
// CHECK-NEXT: [[TMP2:%.*]] = call <8 x half> @llvm.fma.v8f16(<8 x half> [[A:%.*]], <8 x half> [[B:%.*]], <8 x half> [[DOTSPLAT]])
// CHECK-NEXT: ret <8 x half> [[TMP2]]
//
float16x8_t test_vfmasq_n_f16(float16x8_t a, float16x8_t b, float16_t c) {
#ifdef POLYMORPHIC
return vfmasq(a, b, c);
#else /* POLYMORPHIC */
return vfmasq_n_f16(a, b, c);
#endif /* POLYMORPHIC */
}
// CHECK-LABEL: @test_vfmasq_n_f32(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <4 x float> undef, float [[C:%.*]], i32 0
// CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <4 x float> [[DOTSPLATINSERT]], <4 x float> undef, <4 x i32> zeroinitializer
// CHECK-NEXT: [[TMP0:%.*]] = call <4 x float> @llvm.fma.v4f32(<4 x float> [[A:%.*]], <4 x float> [[B:%.*]], <4 x float> [[DOTSPLAT]])
// CHECK-NEXT: ret <4 x float> [[TMP0]]
//
float32x4_t test_vfmasq_n_f32(float32x4_t a, float32x4_t b, float32_t c) {
#ifdef POLYMORPHIC
return vfmasq(a, b, c);
#else /* POLYMORPHIC */
return vfmasq_n_f32(a, b, c);
#endif /* POLYMORPHIC */
}
// CHECK-LABEL: @test_vfmsq_f16(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = fneg <8 x half> [[C:%.*]]
// CHECK-NEXT: [[TMP1:%.*]] = call <8 x half> @llvm.fma.v8f16(<8 x half> [[B:%.*]], <8 x half> [[TMP0]], <8 x half> [[A:%.*]])
// CHECK-NEXT: ret <8 x half> [[TMP1]]
//
float16x8_t test_vfmsq_f16(float16x8_t a, float16x8_t b, float16x8_t c) {
#ifdef POLYMORPHIC
return vfmsq(a, b, c);
#else /* POLYMORPHIC */
return vfmsq_f16(a, b, c);
#endif /* POLYMORPHIC */
}
// CHECK-LABEL: @test_vfmsq_f32(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = fneg <4 x float> [[C:%.*]]
// CHECK-NEXT: [[TMP1:%.*]] = call <4 x float> @llvm.fma.v4f32(<4 x float> [[B:%.*]], <4 x float> [[TMP0]], <4 x float> [[A:%.*]])
// CHECK-NEXT: ret <4 x float> [[TMP1]]
//
float32x4_t test_vfmsq_f32(float32x4_t a, float32x4_t b, float32x4_t c) {
#ifdef POLYMORPHIC
return vfmsq(a, b, c);
#else /* POLYMORPHIC */
return vfmsq_f32(a, b, c);
#endif /* POLYMORPHIC */
}
[ARM,MVE] Add intrinsics and isel for MVE integer VMLA. Summary: These instructions compute multiply+add in integers, with one of the operands being a splat of a scalar. (VMLA and VMLAS differ in whether the splat operand is a multiplier or the addend.) I've represented these in IR using existing standard IR operations for the unpredicated forms. The predicated forms are done with target- specific intrinsics, as usual. When operating on n-bit vector lanes, only the bottom n bits of the i32 scalar operand are used. So we have to tell that to isel lowering, to allow it to remove a pointless sign- or zero-extension instruction on that input register. That's done in `PerformIntrinsicCombine`, but first I had to enable `PerformIntrinsicCombine` for MVE targets (previously all the intrinsics it handled were for NEON), and make it a method of `ARMTargetLowering` so that it can get at `SimplifyDemandedBits`. Reviewers: dmgreen, MarkMurrayARM, miyuki, ostannard Reviewed By: dmgreen Subscribers: kristof.beyls, hiraditya, danielkiss, cfe-commits Tags: #clang Differential Revision: https://reviews.llvm.org/D76122
2020-03-11 20:48:36 +08:00
// CHECK-LABEL: @test_vmlaq_n_s8(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <16 x i8> undef, i8 [[C:%.*]], i32 0
// CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <16 x i8> [[DOTSPLATINSERT]], <16 x i8> undef, <16 x i32> zeroinitializer
// CHECK-NEXT: [[TMP0:%.*]] = mul <16 x i8> [[B:%.*]], [[DOTSPLAT]]
// CHECK-NEXT: [[TMP1:%.*]] = add <16 x i8> [[TMP0]], [[A:%.*]]
// CHECK-NEXT: ret <16 x i8> [[TMP1]]
//
int8x16_t test_vmlaq_n_s8(int8x16_t a, int8x16_t b, int8_t c) {
#ifdef POLYMORPHIC
return vmlaq(a, b, c);
#else /* POLYMORPHIC */
return vmlaq_n_s8(a, b, c);
#endif /* POLYMORPHIC */
}
// CHECK-LABEL: @test_vmlaq_n_s16(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <8 x i16> undef, i16 [[C:%.*]], i32 0
// CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <8 x i16> [[DOTSPLATINSERT]], <8 x i16> undef, <8 x i32> zeroinitializer
// CHECK-NEXT: [[TMP0:%.*]] = mul <8 x i16> [[B:%.*]], [[DOTSPLAT]]
// CHECK-NEXT: [[TMP1:%.*]] = add <8 x i16> [[TMP0]], [[A:%.*]]
// CHECK-NEXT: ret <8 x i16> [[TMP1]]
//
int16x8_t test_vmlaq_n_s16(int16x8_t a, int16x8_t b, int16_t c) {
#ifdef POLYMORPHIC
return vmlaq(a, b, c);
#else /* POLYMORPHIC */
return vmlaq_n_s16(a, b, c);
#endif /* POLYMORPHIC */
}
// CHECK-LABEL: @test_vmlaq_n_s32(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <4 x i32> undef, i32 [[C:%.*]], i32 0
// CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <4 x i32> [[DOTSPLATINSERT]], <4 x i32> undef, <4 x i32> zeroinitializer
// CHECK-NEXT: [[TMP0:%.*]] = mul <4 x i32> [[B:%.*]], [[DOTSPLAT]]
// CHECK-NEXT: [[TMP1:%.*]] = add <4 x i32> [[TMP0]], [[A:%.*]]
// CHECK-NEXT: ret <4 x i32> [[TMP1]]
//
int32x4_t test_vmlaq_n_s32(int32x4_t a, int32x4_t b, int32_t c) {
#ifdef POLYMORPHIC
return vmlaq(a, b, c);
#else /* POLYMORPHIC */
return vmlaq_n_s32(a, b, c);
#endif /* POLYMORPHIC */
}
// CHECK-LABEL: @test_vmlaq_n_u8(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <16 x i8> undef, i8 [[C:%.*]], i32 0
// CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <16 x i8> [[DOTSPLATINSERT]], <16 x i8> undef, <16 x i32> zeroinitializer
// CHECK-NEXT: [[TMP0:%.*]] = mul <16 x i8> [[B:%.*]], [[DOTSPLAT]]
// CHECK-NEXT: [[TMP1:%.*]] = add <16 x i8> [[TMP0]], [[A:%.*]]
// CHECK-NEXT: ret <16 x i8> [[TMP1]]
//
uint8x16_t test_vmlaq_n_u8(uint8x16_t a, uint8x16_t b, uint8_t c) {
#ifdef POLYMORPHIC
return vmlaq(a, b, c);
#else /* POLYMORPHIC */
return vmlaq_n_u8(a, b, c);
#endif /* POLYMORPHIC */
}
// CHECK-LABEL: @test_vmlaq_n_u16(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <8 x i16> undef, i16 [[C:%.*]], i32 0
// CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <8 x i16> [[DOTSPLATINSERT]], <8 x i16> undef, <8 x i32> zeroinitializer
// CHECK-NEXT: [[TMP0:%.*]] = mul <8 x i16> [[B:%.*]], [[DOTSPLAT]]
// CHECK-NEXT: [[TMP1:%.*]] = add <8 x i16> [[TMP0]], [[A:%.*]]
// CHECK-NEXT: ret <8 x i16> [[TMP1]]
//
uint16x8_t test_vmlaq_n_u16(uint16x8_t a, uint16x8_t b, uint16_t c) {
#ifdef POLYMORPHIC
return vmlaq(a, b, c);
#else /* POLYMORPHIC */
return vmlaq_n_u16(a, b, c);
#endif /* POLYMORPHIC */
}
// CHECK-LABEL: @test_vmlaq_n_u32(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <4 x i32> undef, i32 [[C:%.*]], i32 0
// CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <4 x i32> [[DOTSPLATINSERT]], <4 x i32> undef, <4 x i32> zeroinitializer
// CHECK-NEXT: [[TMP0:%.*]] = mul <4 x i32> [[B:%.*]], [[DOTSPLAT]]
// CHECK-NEXT: [[TMP1:%.*]] = add <4 x i32> [[TMP0]], [[A:%.*]]
// CHECK-NEXT: ret <4 x i32> [[TMP1]]
//
uint32x4_t test_vmlaq_n_u32(uint32x4_t a, uint32x4_t b, uint32_t c) {
#ifdef POLYMORPHIC
return vmlaq(a, b, c);
#else /* POLYMORPHIC */
return vmlaq_n_u32(a, b, c);
#endif /* POLYMORPHIC */
}
// CHECK-LABEL: @test_vmlasq_n_s8(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = mul <16 x i8> [[A:%.*]], [[B:%.*]]
// CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <16 x i8> undef, i8 [[C:%.*]], i32 0
// CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <16 x i8> [[DOTSPLATINSERT]], <16 x i8> undef, <16 x i32> zeroinitializer
// CHECK-NEXT: [[TMP1:%.*]] = add <16 x i8> [[TMP0]], [[DOTSPLAT]]
// CHECK-NEXT: ret <16 x i8> [[TMP1]]
//
int8x16_t test_vmlasq_n_s8(int8x16_t a, int8x16_t b, int8_t c) {
#ifdef POLYMORPHIC
return vmlasq(a, b, c);
#else /* POLYMORPHIC */
return vmlasq_n_s8(a, b, c);
#endif /* POLYMORPHIC */
}
// CHECK-LABEL: @test_vmlasq_n_s16(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = mul <8 x i16> [[A:%.*]], [[B:%.*]]
// CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <8 x i16> undef, i16 [[C:%.*]], i32 0
// CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <8 x i16> [[DOTSPLATINSERT]], <8 x i16> undef, <8 x i32> zeroinitializer
// CHECK-NEXT: [[TMP1:%.*]] = add <8 x i16> [[TMP0]], [[DOTSPLAT]]
// CHECK-NEXT: ret <8 x i16> [[TMP1]]
//
int16x8_t test_vmlasq_n_s16(int16x8_t a, int16x8_t b, int16_t c) {
#ifdef POLYMORPHIC
return vmlasq(a, b, c);
#else /* POLYMORPHIC */
return vmlasq_n_s16(a, b, c);
#endif /* POLYMORPHIC */
}
// CHECK-LABEL: @test_vmlasq_n_s32(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = mul <4 x i32> [[A:%.*]], [[B:%.*]]
// CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <4 x i32> undef, i32 [[C:%.*]], i32 0
// CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <4 x i32> [[DOTSPLATINSERT]], <4 x i32> undef, <4 x i32> zeroinitializer
// CHECK-NEXT: [[TMP1:%.*]] = add <4 x i32> [[TMP0]], [[DOTSPLAT]]
// CHECK-NEXT: ret <4 x i32> [[TMP1]]
//
int32x4_t test_vmlasq_n_s32(int32x4_t a, int32x4_t b, int32_t c) {
#ifdef POLYMORPHIC
return vmlasq(a, b, c);
#else /* POLYMORPHIC */
return vmlasq_n_s32(a, b, c);
#endif /* POLYMORPHIC */
}
// CHECK-LABEL: @test_vmlasq_n_u8(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = mul <16 x i8> [[A:%.*]], [[B:%.*]]
// CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <16 x i8> undef, i8 [[C:%.*]], i32 0
// CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <16 x i8> [[DOTSPLATINSERT]], <16 x i8> undef, <16 x i32> zeroinitializer
// CHECK-NEXT: [[TMP1:%.*]] = add <16 x i8> [[TMP0]], [[DOTSPLAT]]
// CHECK-NEXT: ret <16 x i8> [[TMP1]]
//
uint8x16_t test_vmlasq_n_u8(uint8x16_t a, uint8x16_t b, uint8_t c) {
#ifdef POLYMORPHIC
return vmlasq(a, b, c);
#else /* POLYMORPHIC */
return vmlasq_n_u8(a, b, c);
#endif /* POLYMORPHIC */
}
// CHECK-LABEL: @test_vmlasq_n_u16(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = mul <8 x i16> [[A:%.*]], [[B:%.*]]
// CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <8 x i16> undef, i16 [[C:%.*]], i32 0
// CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <8 x i16> [[DOTSPLATINSERT]], <8 x i16> undef, <8 x i32> zeroinitializer
// CHECK-NEXT: [[TMP1:%.*]] = add <8 x i16> [[TMP0]], [[DOTSPLAT]]
// CHECK-NEXT: ret <8 x i16> [[TMP1]]
//
uint16x8_t test_vmlasq_n_u16(uint16x8_t a, uint16x8_t b, uint16_t c) {
#ifdef POLYMORPHIC
return vmlasq(a, b, c);
#else /* POLYMORPHIC */
return vmlasq_n_u16(a, b, c);
#endif /* POLYMORPHIC */
}
// CHECK-LABEL: @test_vmlasq_n_u32(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = mul <4 x i32> [[A:%.*]], [[B:%.*]]
// CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <4 x i32> undef, i32 [[C:%.*]], i32 0
// CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <4 x i32> [[DOTSPLATINSERT]], <4 x i32> undef, <4 x i32> zeroinitializer
// CHECK-NEXT: [[TMP1:%.*]] = add <4 x i32> [[TMP0]], [[DOTSPLAT]]
// CHECK-NEXT: ret <4 x i32> [[TMP1]]
//
uint32x4_t test_vmlasq_n_u32(uint32x4_t a, uint32x4_t b, uint32_t c) {
#ifdef POLYMORPHIC
return vmlasq(a, b, c);
#else /* POLYMORPHIC */
return vmlasq_n_u32(a, b, c);
#endif /* POLYMORPHIC */
}
[ARM,MVE] Add intrinsics and isel for MVE fused multiply-add. Summary: This adds the ACLE intrinsic family for the VFMA and VFMS instructions, which perform fused multiply-add on vectors of floats. I've represented the unpredicated versions in IR using the cross- platform `@llvm.fma` IR intrinsic. We already had isel rules to convert one of those into a vector VFMA in the simplest possible way; but we didn't have rules to detect a negated argument and turn it into VFMS, or rules to detect a splat argument and turn it into one of the two vector/scalar forms of the instruction. Now we have all of those. The predicated form uses a target-specific intrinsic as usual, but I've stuck to just one, for a predicated FMA. The subtraction and splat versions are code-generated by passing an fneg or a splat as one of its operands, the same way as the unpredicated version. In arm_mve_defs.h, I've had to introduce a tiny extra piece of infrastructure: a record `id` for use in codegen dags which implements the identity function. (Just because you can't declare a Tablegen value of type dag which is //only// a `$varname`: you have to wrap it in something. Now I can write `(id $varname)` to get the same effect.) Reviewers: dmgreen, MarkMurrayARM, miyuki, ostannard Reviewed By: dmgreen Subscribers: kristof.beyls, hiraditya, danielkiss, cfe-commits, llvm-commits Tags: #clang, #llvm Differential Revision: https://reviews.llvm.org/D75998
2020-03-12 17:57:48 +08:00
// CHECK-LABEL: @test_vfmaq_m_f16(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
// CHECK-NEXT: [[TMP1:%.*]] = call <8 x i1> @llvm.arm.mve.pred.i2v.v8i1(i32 [[TMP0]])
// CHECK-NEXT: [[TMP2:%.*]] = call <8 x half> @llvm.arm.mve.fma.predicated.v8f16.v8i1(<8 x half> [[B:%.*]], <8 x half> [[C:%.*]], <8 x half> [[A:%.*]], <8 x i1> [[TMP1]])
// CHECK-NEXT: ret <8 x half> [[TMP2]]
//
float16x8_t test_vfmaq_m_f16(float16x8_t a, float16x8_t b, float16x8_t c, mve_pred16_t p) {
#ifdef POLYMORPHIC
return vfmaq_m(a, b, c, p);
#else /* POLYMORPHIC */
return vfmaq_m_f16(a, b, c, p);
#endif /* POLYMORPHIC */
}
// CHECK-LABEL: @test_vfmaq_m_f32(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
// CHECK-NEXT: [[TMP1:%.*]] = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 [[TMP0]])
// CHECK-NEXT: [[TMP2:%.*]] = call <4 x float> @llvm.arm.mve.fma.predicated.v4f32.v4i1(<4 x float> [[B:%.*]], <4 x float> [[C:%.*]], <4 x float> [[A:%.*]], <4 x i1> [[TMP1]])
// CHECK-NEXT: ret <4 x float> [[TMP2]]
//
float32x4_t test_vfmaq_m_f32(float32x4_t a, float32x4_t b, float32x4_t c, mve_pred16_t p) {
#ifdef POLYMORPHIC
return vfmaq_m(a, b, c, p);
#else /* POLYMORPHIC */
return vfmaq_m_f32(a, b, c, p);
#endif /* POLYMORPHIC */
}
// CHECK-LABEL: @test_vfmaq_m_n_f16(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = bitcast float [[C_COERCE:%.*]] to i32
// CHECK-NEXT: [[TMP_0_EXTRACT_TRUNC:%.*]] = trunc i32 [[TMP0]] to i16
// CHECK-NEXT: [[TMP1:%.*]] = bitcast i16 [[TMP_0_EXTRACT_TRUNC]] to half
// CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <8 x half> undef, half [[TMP1]], i32 0
// CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <8 x half> [[DOTSPLATINSERT]], <8 x half> undef, <8 x i32> zeroinitializer
// CHECK-NEXT: [[TMP2:%.*]] = zext i16 [[P:%.*]] to i32
// CHECK-NEXT: [[TMP3:%.*]] = call <8 x i1> @llvm.arm.mve.pred.i2v.v8i1(i32 [[TMP2]])
// CHECK-NEXT: [[TMP4:%.*]] = call <8 x half> @llvm.arm.mve.fma.predicated.v8f16.v8i1(<8 x half> [[B:%.*]], <8 x half> [[DOTSPLAT]], <8 x half> [[A:%.*]], <8 x i1> [[TMP3]])
// CHECK-NEXT: ret <8 x half> [[TMP4]]
//
float16x8_t test_vfmaq_m_n_f16(float16x8_t a, float16x8_t b, float16_t c, mve_pred16_t p) {
#ifdef POLYMORPHIC
return vfmaq_m(a, b, c, p);
#else /* POLYMORPHIC */
return vfmaq_m_n_f16(a, b, c, p);
#endif /* POLYMORPHIC */
}
// CHECK-LABEL: @test_vfmaq_m_n_f32(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <4 x float> undef, float [[C:%.*]], i32 0
// CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <4 x float> [[DOTSPLATINSERT]], <4 x float> undef, <4 x i32> zeroinitializer
// CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
// CHECK-NEXT: [[TMP1:%.*]] = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 [[TMP0]])
// CHECK-NEXT: [[TMP2:%.*]] = call <4 x float> @llvm.arm.mve.fma.predicated.v4f32.v4i1(<4 x float> [[B:%.*]], <4 x float> [[DOTSPLAT]], <4 x float> [[A:%.*]], <4 x i1> [[TMP1]])
// CHECK-NEXT: ret <4 x float> [[TMP2]]
//
float32x4_t test_vfmaq_m_n_f32(float32x4_t a, float32x4_t b, float32_t c, mve_pred16_t p) {
#ifdef POLYMORPHIC
return vfmaq_m(a, b, c, p);
#else /* POLYMORPHIC */
return vfmaq_m_n_f32(a, b, c, p);
#endif /* POLYMORPHIC */
}
// CHECK-LABEL: @test_vfmasq_m_n_f16(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = bitcast float [[C_COERCE:%.*]] to i32
// CHECK-NEXT: [[TMP_0_EXTRACT_TRUNC:%.*]] = trunc i32 [[TMP0]] to i16
// CHECK-NEXT: [[TMP1:%.*]] = bitcast i16 [[TMP_0_EXTRACT_TRUNC]] to half
// CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <8 x half> undef, half [[TMP1]], i32 0
// CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <8 x half> [[DOTSPLATINSERT]], <8 x half> undef, <8 x i32> zeroinitializer
// CHECK-NEXT: [[TMP2:%.*]] = zext i16 [[P:%.*]] to i32
// CHECK-NEXT: [[TMP3:%.*]] = call <8 x i1> @llvm.arm.mve.pred.i2v.v8i1(i32 [[TMP2]])
// CHECK-NEXT: [[TMP4:%.*]] = call <8 x half> @llvm.arm.mve.fma.predicated.v8f16.v8i1(<8 x half> [[A:%.*]], <8 x half> [[B:%.*]], <8 x half> [[DOTSPLAT]], <8 x i1> [[TMP3]])
// CHECK-NEXT: ret <8 x half> [[TMP4]]
//
float16x8_t test_vfmasq_m_n_f16(float16x8_t a, float16x8_t b, float16_t c, mve_pred16_t p) {
#ifdef POLYMORPHIC
return vfmasq_m(a, b, c, p);
#else /* POLYMORPHIC */
return vfmasq_m_n_f16(a, b, c, p);
#endif /* POLYMORPHIC */
}
// CHECK-LABEL: @test_vfmasq_m_n_f32(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <4 x float> undef, float [[C:%.*]], i32 0
// CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <4 x float> [[DOTSPLATINSERT]], <4 x float> undef, <4 x i32> zeroinitializer
// CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
// CHECK-NEXT: [[TMP1:%.*]] = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 [[TMP0]])
// CHECK-NEXT: [[TMP2:%.*]] = call <4 x float> @llvm.arm.mve.fma.predicated.v4f32.v4i1(<4 x float> [[A:%.*]], <4 x float> [[B:%.*]], <4 x float> [[DOTSPLAT]], <4 x i1> [[TMP1]])
// CHECK-NEXT: ret <4 x float> [[TMP2]]
//
float32x4_t test_vfmasq_m_n_f32(float32x4_t a, float32x4_t b, float32_t c, mve_pred16_t p) {
#ifdef POLYMORPHIC
return vfmasq_m(a, b, c, p);
#else /* POLYMORPHIC */
return vfmasq_m_n_f32(a, b, c, p);
#endif /* POLYMORPHIC */
}
// CHECK-LABEL: @test_vfmsq_m_f16(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = fneg <8 x half> [[C:%.*]]
// CHECK-NEXT: [[TMP1:%.*]] = zext i16 [[P:%.*]] to i32
// CHECK-NEXT: [[TMP2:%.*]] = call <8 x i1> @llvm.arm.mve.pred.i2v.v8i1(i32 [[TMP1]])
// CHECK-NEXT: [[TMP3:%.*]] = call <8 x half> @llvm.arm.mve.fma.predicated.v8f16.v8i1(<8 x half> [[B:%.*]], <8 x half> [[TMP0]], <8 x half> [[A:%.*]], <8 x i1> [[TMP2]])
// CHECK-NEXT: ret <8 x half> [[TMP3]]
//
float16x8_t test_vfmsq_m_f16(float16x8_t a, float16x8_t b, float16x8_t c, mve_pred16_t p) {
#ifdef POLYMORPHIC
return vfmsq_m(a, b, c, p);
#else /* POLYMORPHIC */
return vfmsq_m_f16(a, b, c, p);
#endif /* POLYMORPHIC */
}
// CHECK-LABEL: @test_vfmsq_m_f32(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = fneg <4 x float> [[C:%.*]]
// CHECK-NEXT: [[TMP1:%.*]] = zext i16 [[P:%.*]] to i32
// CHECK-NEXT: [[TMP2:%.*]] = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 [[TMP1]])
// CHECK-NEXT: [[TMP3:%.*]] = call <4 x float> @llvm.arm.mve.fma.predicated.v4f32.v4i1(<4 x float> [[B:%.*]], <4 x float> [[TMP0]], <4 x float> [[A:%.*]], <4 x i1> [[TMP2]])
// CHECK-NEXT: ret <4 x float> [[TMP3]]
//
float32x4_t test_vfmsq_m_f32(float32x4_t a, float32x4_t b, float32x4_t c, mve_pred16_t p) {
#ifdef POLYMORPHIC
return vfmsq_m(a, b, c, p);
#else /* POLYMORPHIC */
return vfmsq_m_f32(a, b, c, p);
#endif /* POLYMORPHIC */
}
[ARM,MVE] Add intrinsics and isel for MVE integer VMLA. Summary: These instructions compute multiply+add in integers, with one of the operands being a splat of a scalar. (VMLA and VMLAS differ in whether the splat operand is a multiplier or the addend.) I've represented these in IR using existing standard IR operations for the unpredicated forms. The predicated forms are done with target- specific intrinsics, as usual. When operating on n-bit vector lanes, only the bottom n bits of the i32 scalar operand are used. So we have to tell that to isel lowering, to allow it to remove a pointless sign- or zero-extension instruction on that input register. That's done in `PerformIntrinsicCombine`, but first I had to enable `PerformIntrinsicCombine` for MVE targets (previously all the intrinsics it handled were for NEON), and make it a method of `ARMTargetLowering` so that it can get at `SimplifyDemandedBits`. Reviewers: dmgreen, MarkMurrayARM, miyuki, ostannard Reviewed By: dmgreen Subscribers: kristof.beyls, hiraditya, danielkiss, cfe-commits Tags: #clang Differential Revision: https://reviews.llvm.org/D76122
2020-03-11 20:48:36 +08:00
// CHECK-LABEL: @test_vmlaq_m_n_s8(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = zext i8 [[C:%.*]] to i32
// CHECK-NEXT: [[TMP1:%.*]] = zext i16 [[P:%.*]] to i32
// CHECK-NEXT: [[TMP2:%.*]] = call <16 x i1> @llvm.arm.mve.pred.i2v.v16i1(i32 [[TMP1]])
// CHECK-NEXT: [[TMP3:%.*]] = call <16 x i8> @llvm.arm.mve.vmla.n.predicated.v16i8.v16i1(<16 x i8> [[B:%.*]], <16 x i8> [[A:%.*]], i32 [[TMP0]], <16 x i1> [[TMP2]])
// CHECK-NEXT: ret <16 x i8> [[TMP3]]
//
int8x16_t test_vmlaq_m_n_s8(int8x16_t a, int8x16_t b, int8_t c, mve_pred16_t p) {
#ifdef POLYMORPHIC
return vmlaq_m(a, b, c, p);
#else /* POLYMORPHIC */
return vmlaq_m_n_s8(a, b, c, p);
#endif /* POLYMORPHIC */
}
// CHECK-LABEL: @test_vmlaq_m_n_s16(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[C:%.*]] to i32
// CHECK-NEXT: [[TMP1:%.*]] = zext i16 [[P:%.*]] to i32
// CHECK-NEXT: [[TMP2:%.*]] = call <8 x i1> @llvm.arm.mve.pred.i2v.v8i1(i32 [[TMP1]])
// CHECK-NEXT: [[TMP3:%.*]] = call <8 x i16> @llvm.arm.mve.vmla.n.predicated.v8i16.v8i1(<8 x i16> [[B:%.*]], <8 x i16> [[A:%.*]], i32 [[TMP0]], <8 x i1> [[TMP2]])
// CHECK-NEXT: ret <8 x i16> [[TMP3]]
//
int16x8_t test_vmlaq_m_n_s16(int16x8_t a, int16x8_t b, int16_t c, mve_pred16_t p) {
#ifdef POLYMORPHIC
return vmlaq_m(a, b, c, p);
#else /* POLYMORPHIC */
return vmlaq_m_n_s16(a, b, c, p);
#endif /* POLYMORPHIC */
}
// CHECK-LABEL: @test_vmlaq_m_n_s32(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
// CHECK-NEXT: [[TMP1:%.*]] = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 [[TMP0]])
// CHECK-NEXT: [[TMP2:%.*]] = call <4 x i32> @llvm.arm.mve.vmla.n.predicated.v4i32.v4i1(<4 x i32> [[B:%.*]], <4 x i32> [[A:%.*]], i32 [[C:%.*]], <4 x i1> [[TMP1]])
// CHECK-NEXT: ret <4 x i32> [[TMP2]]
//
int32x4_t test_vmlaq_m_n_s32(int32x4_t a, int32x4_t b, int32_t c, mve_pred16_t p) {
#ifdef POLYMORPHIC
return vmlaq_m(a, b, c, p);
#else /* POLYMORPHIC */
return vmlaq_m_n_s32(a, b, c, p);
#endif /* POLYMORPHIC */
}
// CHECK-LABEL: @test_vmlaq_m_n_u8(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = zext i8 [[C:%.*]] to i32
// CHECK-NEXT: [[TMP1:%.*]] = zext i16 [[P:%.*]] to i32
// CHECK-NEXT: [[TMP2:%.*]] = call <16 x i1> @llvm.arm.mve.pred.i2v.v16i1(i32 [[TMP1]])
// CHECK-NEXT: [[TMP3:%.*]] = call <16 x i8> @llvm.arm.mve.vmla.n.predicated.v16i8.v16i1(<16 x i8> [[B:%.*]], <16 x i8> [[A:%.*]], i32 [[TMP0]], <16 x i1> [[TMP2]])
// CHECK-NEXT: ret <16 x i8> [[TMP3]]
//
uint8x16_t test_vmlaq_m_n_u8(uint8x16_t a, uint8x16_t b, uint8_t c, mve_pred16_t p) {
#ifdef POLYMORPHIC
return vmlaq_m(a, b, c, p);
#else /* POLYMORPHIC */
return vmlaq_m_n_u8(a, b, c, p);
#endif /* POLYMORPHIC */
}
// CHECK-LABEL: @test_vmlaq_m_n_u16(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[C:%.*]] to i32
// CHECK-NEXT: [[TMP1:%.*]] = zext i16 [[P:%.*]] to i32
// CHECK-NEXT: [[TMP2:%.*]] = call <8 x i1> @llvm.arm.mve.pred.i2v.v8i1(i32 [[TMP1]])
// CHECK-NEXT: [[TMP3:%.*]] = call <8 x i16> @llvm.arm.mve.vmla.n.predicated.v8i16.v8i1(<8 x i16> [[B:%.*]], <8 x i16> [[A:%.*]], i32 [[TMP0]], <8 x i1> [[TMP2]])
// CHECK-NEXT: ret <8 x i16> [[TMP3]]
//
uint16x8_t test_vmlaq_m_n_u16(uint16x8_t a, uint16x8_t b, uint16_t c, mve_pred16_t p) {
#ifdef POLYMORPHIC
return vmlaq_m(a, b, c, p);
#else /* POLYMORPHIC */
return vmlaq_m_n_u16(a, b, c, p);
#endif /* POLYMORPHIC */
}
// CHECK-LABEL: @test_vmlaq_m_n_u32(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
// CHECK-NEXT: [[TMP1:%.*]] = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 [[TMP0]])
// CHECK-NEXT: [[TMP2:%.*]] = call <4 x i32> @llvm.arm.mve.vmla.n.predicated.v4i32.v4i1(<4 x i32> [[B:%.*]], <4 x i32> [[A:%.*]], i32 [[C:%.*]], <4 x i1> [[TMP1]])
// CHECK-NEXT: ret <4 x i32> [[TMP2]]
//
uint32x4_t test_vmlaq_m_n_u32(uint32x4_t a, uint32x4_t b, uint32_t c, mve_pred16_t p) {
#ifdef POLYMORPHIC
return vmlaq_m(a, b, c, p);
#else /* POLYMORPHIC */
return vmlaq_m_n_u32(a, b, c, p);
#endif /* POLYMORPHIC */
}
// CHECK-LABEL: @test_vmlasq_m_n_s8(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = zext i8 [[C:%.*]] to i32
// CHECK-NEXT: [[TMP1:%.*]] = zext i16 [[P:%.*]] to i32
// CHECK-NEXT: [[TMP2:%.*]] = call <16 x i1> @llvm.arm.mve.pred.i2v.v16i1(i32 [[TMP1]])
// CHECK-NEXT: [[TMP3:%.*]] = call <16 x i8> @llvm.arm.mve.vmlas.n.predicated.v16i8.v16i1(<16 x i8> [[A:%.*]], <16 x i8> [[B:%.*]], i32 [[TMP0]], <16 x i1> [[TMP2]])
// CHECK-NEXT: ret <16 x i8> [[TMP3]]
//
int8x16_t test_vmlasq_m_n_s8(int8x16_t a, int8x16_t b, int8_t c, mve_pred16_t p) {
#ifdef POLYMORPHIC
return vmlasq_m(a, b, c, p);
#else /* POLYMORPHIC */
return vmlasq_m_n_s8(a, b, c, p);
#endif /* POLYMORPHIC */
}
// CHECK-LABEL: @test_vmlasq_m_n_s16(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[C:%.*]] to i32
// CHECK-NEXT: [[TMP1:%.*]] = zext i16 [[P:%.*]] to i32
// CHECK-NEXT: [[TMP2:%.*]] = call <8 x i1> @llvm.arm.mve.pred.i2v.v8i1(i32 [[TMP1]])
// CHECK-NEXT: [[TMP3:%.*]] = call <8 x i16> @llvm.arm.mve.vmlas.n.predicated.v8i16.v8i1(<8 x i16> [[A:%.*]], <8 x i16> [[B:%.*]], i32 [[TMP0]], <8 x i1> [[TMP2]])
// CHECK-NEXT: ret <8 x i16> [[TMP3]]
//
int16x8_t test_vmlasq_m_n_s16(int16x8_t a, int16x8_t b, int16_t c, mve_pred16_t p) {
#ifdef POLYMORPHIC
return vmlasq_m(a, b, c, p);
#else /* POLYMORPHIC */
return vmlasq_m_n_s16(a, b, c, p);
#endif /* POLYMORPHIC */
}
// CHECK-LABEL: @test_vmlasq_m_n_s32(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
// CHECK-NEXT: [[TMP1:%.*]] = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 [[TMP0]])
// CHECK-NEXT: [[TMP2:%.*]] = call <4 x i32> @llvm.arm.mve.vmlas.n.predicated.v4i32.v4i1(<4 x i32> [[A:%.*]], <4 x i32> [[B:%.*]], i32 [[C:%.*]], <4 x i1> [[TMP1]])
// CHECK-NEXT: ret <4 x i32> [[TMP2]]
//
int32x4_t test_vmlasq_m_n_s32(int32x4_t a, int32x4_t b, int32_t c, mve_pred16_t p) {
#ifdef POLYMORPHIC
return vmlasq_m(a, b, c, p);
#else /* POLYMORPHIC */
return vmlasq_m_n_s32(a, b, c, p);
#endif /* POLYMORPHIC */
}
// CHECK-LABEL: @test_vmlasq_m_n_u8(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = zext i8 [[C:%.*]] to i32
// CHECK-NEXT: [[TMP1:%.*]] = zext i16 [[P:%.*]] to i32
// CHECK-NEXT: [[TMP2:%.*]] = call <16 x i1> @llvm.arm.mve.pred.i2v.v16i1(i32 [[TMP1]])
// CHECK-NEXT: [[TMP3:%.*]] = call <16 x i8> @llvm.arm.mve.vmlas.n.predicated.v16i8.v16i1(<16 x i8> [[A:%.*]], <16 x i8> [[B:%.*]], i32 [[TMP0]], <16 x i1> [[TMP2]])
// CHECK-NEXT: ret <16 x i8> [[TMP3]]
//
uint8x16_t test_vmlasq_m_n_u8(uint8x16_t a, uint8x16_t b, uint8_t c, mve_pred16_t p) {
#ifdef POLYMORPHIC
return vmlasq_m(a, b, c, p);
#else /* POLYMORPHIC */
return vmlasq_m_n_u8(a, b, c, p);
#endif /* POLYMORPHIC */
}
// CHECK-LABEL: @test_vmlasq_m_n_u16(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[C:%.*]] to i32
// CHECK-NEXT: [[TMP1:%.*]] = zext i16 [[P:%.*]] to i32
// CHECK-NEXT: [[TMP2:%.*]] = call <8 x i1> @llvm.arm.mve.pred.i2v.v8i1(i32 [[TMP1]])
// CHECK-NEXT: [[TMP3:%.*]] = call <8 x i16> @llvm.arm.mve.vmlas.n.predicated.v8i16.v8i1(<8 x i16> [[A:%.*]], <8 x i16> [[B:%.*]], i32 [[TMP0]], <8 x i1> [[TMP2]])
// CHECK-NEXT: ret <8 x i16> [[TMP3]]
//
uint16x8_t test_vmlasq_m_n_u16(uint16x8_t a, uint16x8_t b, uint16_t c, mve_pred16_t p) {
#ifdef POLYMORPHIC
return vmlasq_m(a, b, c, p);
#else /* POLYMORPHIC */
return vmlasq_m_n_u16(a, b, c, p);
#endif /* POLYMORPHIC */
}
// CHECK-LABEL: @test_vmlasq_m_n_u32(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = zext i16 [[P:%.*]] to i32
// CHECK-NEXT: [[TMP1:%.*]] = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 [[TMP0]])
// CHECK-NEXT: [[TMP2:%.*]] = call <4 x i32> @llvm.arm.mve.vmlas.n.predicated.v4i32.v4i1(<4 x i32> [[A:%.*]], <4 x i32> [[B:%.*]], i32 [[C:%.*]], <4 x i1> [[TMP1]])
// CHECK-NEXT: ret <4 x i32> [[TMP2]]
//
uint32x4_t test_vmlasq_m_n_u32(uint32x4_t a, uint32x4_t b, uint32_t c, mve_pred16_t p) {
#ifdef POLYMORPHIC
return vmlasq_m(a, b, c, p);
#else /* POLYMORPHIC */
return vmlasq_m_n_u32(a, b, c, p);
#endif /* POLYMORPHIC */
}