[clang] [CIR][AArch64] Upstream pairwise-minimum NEON builtins (PR #191759)
Andrzej Warzyński via cfe-commits
cfe-commits at lists.llvm.org
Thu Apr 16 04:28:18 PDT 2026
================
@@ -1778,6 +1778,326 @@ float64x2_t test_vminnmq_f64(float64x2_t v1, float64x2_t v2) {
return vminnmq_f64(v1, v2);
}
+//===----------------------------------------------------------------------===//
+// 2.1.8.6 Pairwise minimum
+// https://arm-software.github.io/acle/neon_intrinsics/advsimd.html#pairwise-minimum
+//===----------------------------------------------------------------------===//
+
+// LLVM-LABEL: @test_vpmin_s8(
+// CIR-LABEL: @vpmin_s8(
+int8x8_t test_vpmin_s8(int8x8_t a, int8x8_t b) {
+// CIR: cir.call_llvm_intrinsic "aarch64.neon.sminp" %{{.*}}, %{{.*}} : (!cir.vector<8 x !s8i>, !cir.vector<8 x !s8i>) -> !cir.vector<8 x !s8i>
+
+// LLVM-SAME: <8 x i8> noundef [[A:%.*]], <8 x i8> noundef [[B:%.*]])
+// LLVM: [[VPMIN_I:%.*]] = call <8 x i8> @llvm.aarch64.neon.sminp.v8i8(<8 x i8> [[A]], <8 x i8> [[B]])
+// LLVM-NEXT: ret <8 x i8> [[VPMIN_I]]
+ return vpmin_s8(a, b);
+}
+
+// LLVM-LABEL: @test_vpmin_s16(
+// CIR-LABEL: @vpmin_s16(
+int16x4_t test_vpmin_s16(int16x4_t a, int16x4_t b) {
+// CIR: cir.call_llvm_intrinsic "aarch64.neon.sminp" %{{.*}}, %{{.*}} : (!cir.vector<4 x !s16i>, !cir.vector<4 x !s16i>) -> !cir.vector<4 x !s16i>
+
+// LLVM-SAME: <4 x i16> noundef [[A:%.*]], <4 x i16> noundef [[B:%.*]])
+// LLVM: [[TMP0:%.*]] = bitcast <4 x i16> [[A]] to <8 x i8>
+// LLVM-NEXT: [[TMP1:%.*]] = bitcast <4 x i16> [[B]] to <8 x i8>
+// LLVM-NEXT: [[VPMIN_I:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// LLVM-NEXT: [[VPMIN1_I:%.*]] = bitcast <8 x i8> [[TMP1]] to <4 x i16>
+// LLVM-NEXT: [[VPMIN2_I:%.*]] = call <4 x i16> @llvm.aarch64.neon.sminp.v4i16(<4 x i16> [[VPMIN_I]], <4 x i16> [[VPMIN1_I]])
+// LLVM-NEXT: ret <4 x i16> [[VPMIN2_I]]
+ return vpmin_s16(a, b);
+}
+
+// LLVM-LABEL: @test_vpmin_s32(
+// CIR-LABEL: @vpmin_s32(
+int32x2_t test_vpmin_s32(int32x2_t a, int32x2_t b) {
+// CIR: cir.call_llvm_intrinsic "aarch64.neon.sminp" %{{.*}}, %{{.*}} : (!cir.vector<2 x !s32i>, !cir.vector<2 x !s32i>) -> !cir.vector<2 x !s32i>
+
+// LLVM-SAME: <2 x i32> noundef [[A:%.*]], <2 x i32> noundef [[B:%.*]])
+// LLVM: [[TMP0:%.*]] = bitcast <2 x i32> [[A]] to <8 x i8>
+// LLVM-NEXT: [[TMP1:%.*]] = bitcast <2 x i32> [[B]] to <8 x i8>
+// LLVM-NEXT: [[VPMIN_I:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// LLVM-NEXT: [[VPMIN1_I:%.*]] = bitcast <8 x i8> [[TMP1]] to <2 x i32>
+// LLVM-NEXT: [[VPMIN2_I:%.*]] = call <2 x i32> @llvm.aarch64.neon.sminp.v2i32(<2 x i32> [[VPMIN_I]], <2 x i32> [[VPMIN1_I]])
+// LLVM-NEXT: ret <2 x i32> [[VPMIN2_I]]
+ return vpmin_s32(a, b);
+}
+
+// LLVM-LABEL: @test_vpmin_u8(
+// CIR-LABEL: @vpmin_u8(
+uint8x8_t test_vpmin_u8(uint8x8_t a, uint8x8_t b) {
+// CIR: cir.call_llvm_intrinsic "aarch64.neon.uminp" %{{.*}}, %{{.*}} : (!cir.vector<8 x !u8i>, !cir.vector<8 x !u8i>) -> !cir.vector<8 x !u8i>
+
+// LLVM-SAME: <8 x i8> noundef [[A:%.*]], <8 x i8> noundef [[B:%.*]])
+// LLVM: [[VPMIN_I:%.*]] = call <8 x i8> @llvm.aarch64.neon.uminp.v8i8(<8 x i8> [[A]], <8 x i8> [[B]])
+// LLVM-NEXT: ret <8 x i8> [[VPMIN_I]]
+ return vpmin_u8(a, b);
+}
+
+// LLVM-LABEL: @test_vpmin_u16(
+// CIR-LABEL: @vpmin_u16(
+uint16x4_t test_vpmin_u16(uint16x4_t a, uint16x4_t b) {
+// CIR: cir.call_llvm_intrinsic "aarch64.neon.uminp" %{{.*}}, %{{.*}} : (!cir.vector<4 x !u16i>, !cir.vector<4 x !u16i>) -> !cir.vector<4 x !u16i>
+
+// LLVM-SAME: <4 x i16> noundef [[A:%.*]], <4 x i16> noundef [[B:%.*]])
+// LLVM: [[TMP0:%.*]] = bitcast <4 x i16> [[A]] to <8 x i8>
+// LLVM-NEXT: [[TMP1:%.*]] = bitcast <4 x i16> [[B]] to <8 x i8>
+// LLVM-NEXT: [[VPMIN_I:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// LLVM-NEXT: [[VPMIN1_I:%.*]] = bitcast <8 x i8> [[TMP1]] to <4 x i16>
+// LLVM-NEXT: [[VPMIN2_I:%.*]] = call <4 x i16> @llvm.aarch64.neon.uminp.v4i16(<4 x i16> [[VPMIN_I]], <4 x i16> [[VPMIN1_I]])
+// LLVM-NEXT: ret <4 x i16> [[VPMIN2_I]]
+ return vpmin_u16(a, b);
+}
+
+// LLVM-LABEL: @test_vpmin_u32(
+// CIR-LABEL: @vpmin_u32(
+uint32x2_t test_vpmin_u32(uint32x2_t a, uint32x2_t b) {
+// CIR: cir.call_llvm_intrinsic "aarch64.neon.uminp" %{{.*}}, %{{.*}} : (!cir.vector<2 x !u32i>, !cir.vector<2 x !u32i>) -> !cir.vector<2 x !u32i>
+
+// LLVM-SAME: <2 x i32> noundef [[A:%.*]], <2 x i32> noundef [[B:%.*]])
+// LLVM: [[TMP0:%.*]] = bitcast <2 x i32> [[A]] to <8 x i8>
+// LLVM-NEXT: [[TMP1:%.*]] = bitcast <2 x i32> [[B]] to <8 x i8>
+// LLVM-NEXT: [[VPMIN_I:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// LLVM-NEXT: [[VPMIN1_I:%.*]] = bitcast <8 x i8> [[TMP1]] to <2 x i32>
+// LLVM-NEXT: [[VPMIN2_I:%.*]] = call <2 x i32> @llvm.aarch64.neon.uminp.v2i32(<2 x i32> [[VPMIN_I]], <2 x i32> [[VPMIN1_I]])
+// LLVM-NEXT: ret <2 x i32> [[VPMIN2_I]]
+ return vpmin_u32(a, b);
+}
+
+// LLVM-LABEL: @test_vpmin_f32(
+// CIR-LABEL: @vpmin_f32(
+float32x2_t test_vpmin_f32(float32x2_t a, float32x2_t b) {
+// CIR: cir.call_llvm_intrinsic "aarch64.neon.fminp" %{{.*}}, %{{.*}} : (!cir.vector<2 x !cir.float>, !cir.vector<2 x !cir.float>) -> !cir.vector<2 x !cir.float>
+
+// LLVM-SAME: <2 x float> noundef [[A:%.*]], <2 x float> noundef [[B:%.*]])
+// LLVM: [[TMP0:%.*]] = bitcast <2 x float> [[A]] to <2 x i32>
+// LLVM-NEXT: [[TMP1:%.*]] = bitcast <2 x float> [[B]] to <2 x i32>
+// LLVM-NEXT: [[TMP2:%.*]] = bitcast <2 x i32> [[TMP0]] to <8 x i8>
+// LLVM-NEXT: [[TMP3:%.*]] = bitcast <2 x i32> [[TMP1]] to <8 x i8>
+// LLVM-NEXT: [[VPMIN_I:%.*]] = bitcast <8 x i8> [[TMP2]] to <2 x float>
+// LLVM-NEXT: [[VPMIN1_I:%.*]] = bitcast <8 x i8> [[TMP3]] to <2 x float>
+// LLVM-NEXT: [[VPMIN2_I:%.*]] = call <2 x float> @llvm.aarch64.neon.fminp.v2f32(<2 x float> [[VPMIN_I]], <2 x float> [[VPMIN1_I]])
+// LLVM-NEXT: ret <2 x float> [[VPMIN2_I]]
+ return vpmin_f32(a, b);
+}
+
+// LLVM-LABEL: @test_vpminq_s8(
+// CIR-LABEL: @vpminq_s8(
+int8x16_t test_vpminq_s8(int8x16_t a, int8x16_t b) {
+// CIR: cir.call_llvm_intrinsic "aarch64.neon.sminp" %{{.*}}, %{{.*}} : (!cir.vector<16 x !s8i>, !cir.vector<16 x !s8i>) -> !cir.vector<16 x !s8i>
+
+// LLVM-SAME: <16 x i8> noundef [[A:%.*]], <16 x i8> noundef [[B:%.*]])
+// LLVM: [[VPMIN_I:%.*]] = call <16 x i8> @llvm.aarch64.neon.sminp.v16i8(<16 x i8> [[A]], <16 x i8> [[B]])
+// LLVM-NEXT: ret <16 x i8> [[VPMIN_I]]
+ return vpminq_s8(a, b);
+}
+
+// LLVM-LABEL: @test_vpminq_s16(
+// CIR-LABEL: @vpminq_s16(
+int16x8_t test_vpminq_s16(int16x8_t a, int16x8_t b) {
+// CIR: cir.call_llvm_intrinsic "aarch64.neon.sminp" %{{.*}}, %{{.*}} : (!cir.vector<8 x !s16i>, !cir.vector<8 x !s16i>) -> !cir.vector<8 x !s16i>
+
+// LLVM-SAME: <8 x i16> noundef [[A:%.*]], <8 x i16> noundef [[B:%.*]])
+// LLVM: [[TMP0:%.*]] = bitcast <8 x i16> [[A]] to <16 x i8>
+// LLVM-NEXT: [[TMP1:%.*]] = bitcast <8 x i16> [[B]] to <16 x i8>
+// LLVM-NEXT: [[VPMIN_I:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// LLVM-NEXT: [[VPMIN1_I:%.*]] = bitcast <16 x i8> [[TMP1]] to <8 x i16>
+// LLVM-NEXT: [[VPMIN2_I:%.*]] = call <8 x i16> @llvm.aarch64.neon.sminp.v8i16(<8 x i16> [[VPMIN_I]], <8 x i16> [[VPMIN1_I]])
+// LLVM-NEXT: ret <8 x i16> [[VPMIN2_I]]
+ return vpminq_s16(a, b);
+}
+
+// LLVM-LABEL: @test_vpminq_s32(
+// CIR-LABEL: @vpminq_s32(
+int32x4_t test_vpminq_s32(int32x4_t a, int32x4_t b) {
+// CIR: cir.call_llvm_intrinsic "aarch64.neon.sminp" %{{.*}}, %{{.*}} : (!cir.vector<4 x !s32i>, !cir.vector<4 x !s32i>) -> !cir.vector<4 x !s32i>
+
+// LLVM-SAME: <4 x i32> noundef [[A:%.*]], <4 x i32> noundef [[B:%.*]])
+// LLVM: [[TMP0:%.*]] = bitcast <4 x i32> [[A]] to <16 x i8>
+// LLVM-NEXT: [[TMP1:%.*]] = bitcast <4 x i32> [[B]] to <16 x i8>
+// LLVM-NEXT: [[VPMIN_I:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// LLVM-NEXT: [[VPMIN1_I:%.*]] = bitcast <16 x i8> [[TMP1]] to <4 x i32>
+// LLVM-NEXT: [[VPMIN2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sminp.v4i32(<4 x i32> [[VPMIN_I]], <4 x i32> [[VPMIN1_I]])
+// LLVM-NEXT: ret <4 x i32> [[VPMIN2_I]]
+ return vpminq_s32(a, b);
+}
+
+// LLVM-LABEL: @test_vpminq_u8(
+// CIR-LABEL: @vpminq_u8(
+uint8x16_t test_vpminq_u8(uint8x16_t a, uint8x16_t b) {
+// CIR: cir.call_llvm_intrinsic "aarch64.neon.uminp" %{{.*}}, %{{.*}} : (!cir.vector<16 x !u8i>, !cir.vector<16 x !u8i>) -> !cir.vector<16 x !u8i>
+
+// LLVM-SAME: <16 x i8> noundef [[A:%.*]], <16 x i8> noundef [[B:%.*]])
+// LLVM: [[VPMIN_I:%.*]] = call <16 x i8> @llvm.aarch64.neon.uminp.v16i8(<16 x i8> [[A]], <16 x i8> [[B]])
+// LLVM-NEXT: ret <16 x i8> [[VPMIN_I]]
+ return vpminq_u8(a, b);
+}
+
+// LLVM-LABEL: @test_vpminq_u16(
+// CIR-LABEL: @vpminq_u16(
+uint16x8_t test_vpminq_u16(uint16x8_t a, uint16x8_t b) {
+// CIR: cir.call_llvm_intrinsic "aarch64.neon.uminp" %{{.*}}, %{{.*}} : (!cir.vector<8 x !u16i>, !cir.vector<8 x !u16i>) -> !cir.vector<8 x !u16i>
+
+// LLVM-SAME: <8 x i16> noundef [[A:%.*]], <8 x i16> noundef [[B:%.*]])
----------------
banach-space wrote:
Here and in other places, no need to check for the attributes:
```suggestion
// LLVM-SAME: <8 x i16> {{.*}} [[A:%.*]], <8 x i16> {{.*}} [[B:%.*]])
```
https://github.com/llvm/llvm-project/pull/191759
More information about the cfe-commits
mailing list