[clang] [llvm] [Clang][AArch64] Allow FP8 Neon vector types to be used by __builtin_shufflevector (PR #119031)

via llvm-commits llvm-commits at lists.llvm.org
Fri Jan 10 01:52:22 PST 2025


================
@@ -0,0 +1,123 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --version 5
+// RUN: %clang_cc1 -triple aarch64-linux -target-feature +neon -disable-O0-optnone -emit-llvm -o - %s | opt -S -passes=mem2reg | FileCheck %s
+
+// REQUIRES: aarch64-registered-target
+
+typedef __attribute__((neon_vector_type(8))) signed char int8x8_t;
+typedef __attribute__((neon_vector_type(16))) signed char int8x16_t;
+
+typedef __MFloat8x8_t mfloat8x8_t;
+typedef __MFloat8x16_t mfloat8x16_t;
+
+// CHECK-LABEL: define dso_local <8 x i8> @f0(
+// CHECK-SAME: <8 x i8> [[X:%.*]]) #[[ATTR0:[0-9]+]] {
+// CHECK-NEXT:  [[ENTRY:.*:]]
+// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i8> [[X]], <8 x i8> [[X]], <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 3, i32 2, i32 1, i32 0>
+// CHECK-NEXT:    ret <8 x i8> [[SHUFFLE]]
+//
+mfloat8x8_t f0(mfloat8x8_t x) {
+  return __builtin_shufflevector(x, x, 3, 2, 1, 0, 3, 2, 1, 0);
+}
+
+// CHECK-LABEL: define dso_local <8 x i8> @f1(
+// CHECK-SAME: <8 x i8> [[X:%.*]], <8 x i8> noundef [[P:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT:  [[ENTRY:.*:]]
+// CHECK-NEXT:    [[MASK:%.*]] = and <8 x i8> [[P]], splat (i8 7)
+// CHECK-NEXT:    [[SHUF_IDX:%.*]] = extractelement <8 x i8> [[MASK]], i64 0
+// CHECK-NEXT:    [[SHUF_ELT:%.*]] = extractelement <8 x i8> [[X]], i8 [[SHUF_IDX]]
+// CHECK-NEXT:    [[SHUF_INS:%.*]] = insertelement <8 x i8> poison, i8 [[SHUF_ELT]], i64 0
+// CHECK-NEXT:    [[SHUF_IDX1:%.*]] = extractelement <8 x i8> [[MASK]], i64 1
+// CHECK-NEXT:    [[SHUF_ELT2:%.*]] = extractelement <8 x i8> [[X]], i8 [[SHUF_IDX1]]
+// CHECK-NEXT:    [[SHUF_INS3:%.*]] = insertelement <8 x i8> [[SHUF_INS]], i8 [[SHUF_ELT2]], i64 1
+// CHECK-NEXT:    [[SHUF_IDX4:%.*]] = extractelement <8 x i8> [[MASK]], i64 2
+// CHECK-NEXT:    [[SHUF_ELT5:%.*]] = extractelement <8 x i8> [[X]], i8 [[SHUF_IDX4]]
+// CHECK-NEXT:    [[SHUF_INS6:%.*]] = insertelement <8 x i8> [[SHUF_INS3]], i8 [[SHUF_ELT5]], i64 2
+// CHECK-NEXT:    [[SHUF_IDX7:%.*]] = extractelement <8 x i8> [[MASK]], i64 3
+// CHECK-NEXT:    [[SHUF_ELT8:%.*]] = extractelement <8 x i8> [[X]], i8 [[SHUF_IDX7]]
+// CHECK-NEXT:    [[SHUF_INS9:%.*]] = insertelement <8 x i8> [[SHUF_INS6]], i8 [[SHUF_ELT8]], i64 3
+// CHECK-NEXT:    [[SHUF_IDX10:%.*]] = extractelement <8 x i8> [[MASK]], i64 4
+// CHECK-NEXT:    [[SHUF_ELT11:%.*]] = extractelement <8 x i8> [[X]], i8 [[SHUF_IDX10]]
+// CHECK-NEXT:    [[SHUF_INS12:%.*]] = insertelement <8 x i8> [[SHUF_INS9]], i8 [[SHUF_ELT11]], i64 4
+// CHECK-NEXT:    [[SHUF_IDX13:%.*]] = extractelement <8 x i8> [[MASK]], i64 5
+// CHECK-NEXT:    [[SHUF_ELT14:%.*]] = extractelement <8 x i8> [[X]], i8 [[SHUF_IDX13]]
+// CHECK-NEXT:    [[SHUF_INS15:%.*]] = insertelement <8 x i8> [[SHUF_INS12]], i8 [[SHUF_ELT14]], i64 5
+// CHECK-NEXT:    [[SHUF_IDX16:%.*]] = extractelement <8 x i8> [[MASK]], i64 6
+// CHECK-NEXT:    [[SHUF_ELT17:%.*]] = extractelement <8 x i8> [[X]], i8 [[SHUF_IDX16]]
+// CHECK-NEXT:    [[SHUF_INS18:%.*]] = insertelement <8 x i8> [[SHUF_INS15]], i8 [[SHUF_ELT17]], i64 6
+// CHECK-NEXT:    [[SHUF_IDX19:%.*]] = extractelement <8 x i8> [[MASK]], i64 7
+// CHECK-NEXT:    [[SHUF_ELT20:%.*]] = extractelement <8 x i8> [[X]], i8 [[SHUF_IDX19]]
+// CHECK-NEXT:    [[SHUF_INS21:%.*]] = insertelement <8 x i8> [[SHUF_INS18]], i8 [[SHUF_ELT20]], i64 7
+// CHECK-NEXT:    ret <8 x i8> [[SHUF_INS21]]
+//
+mfloat8x8_t f1(mfloat8x8_t x, int8x8_t p) {
+  return __builtin_shufflevector(x, p);
+}
+
+// CHECK-LABEL: define dso_local <16 x i8> @f3(
+// CHECK-SAME: <16 x i8> [[X:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT:  [[ENTRY:.*:]]
+// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <16 x i8> [[X]], <16 x i8> [[X]], <16 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+// CHECK-NEXT:    ret <16 x i8> [[SHUFFLE]]
+//
+mfloat8x16_t f3(mfloat8x16_t x) {
+  return __builtin_shufflevector(x, x, 7, 6, 5, 4, 3, 2, 1, 0, 7, 6, 5, 4, 3, 2,
+                                 1, 0);
+}
+
+// CHECK-LABEL: define dso_local <16 x i8> @f4(
+// CHECK-SAME: <16 x i8> [[X:%.*]], <16 x i8> noundef [[P:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT:  [[ENTRY:.*:]]
+// CHECK-NEXT:    [[MASK:%.*]] = and <16 x i8> [[P]], splat (i8 15)
+// CHECK-NEXT:    [[SHUF_IDX:%.*]] = extractelement <16 x i8> [[MASK]], i64 0
+// CHECK-NEXT:    [[SHUF_ELT:%.*]] = extractelement <16 x i8> [[X]], i8 [[SHUF_IDX]]
+// CHECK-NEXT:    [[SHUF_INS:%.*]] = insertelement <16 x i8> poison, i8 [[SHUF_ELT]], i64 0
+// CHECK-NEXT:    [[SHUF_IDX1:%.*]] = extractelement <16 x i8> [[MASK]], i64 1
+// CHECK-NEXT:    [[SHUF_ELT2:%.*]] = extractelement <16 x i8> [[X]], i8 [[SHUF_IDX1]]
+// CHECK-NEXT:    [[SHUF_INS3:%.*]] = insertelement <16 x i8> [[SHUF_INS]], i8 [[SHUF_ELT2]], i64 1
+// CHECK-NEXT:    [[SHUF_IDX4:%.*]] = extractelement <16 x i8> [[MASK]], i64 2
+// CHECK-NEXT:    [[SHUF_ELT5:%.*]] = extractelement <16 x i8> [[X]], i8 [[SHUF_IDX4]]
+// CHECK-NEXT:    [[SHUF_INS6:%.*]] = insertelement <16 x i8> [[SHUF_INS3]], i8 [[SHUF_ELT5]], i64 2
+// CHECK-NEXT:    [[SHUF_IDX7:%.*]] = extractelement <16 x i8> [[MASK]], i64 3
+// CHECK-NEXT:    [[SHUF_ELT8:%.*]] = extractelement <16 x i8> [[X]], i8 [[SHUF_IDX7]]
+// CHECK-NEXT:    [[SHUF_INS9:%.*]] = insertelement <16 x i8> [[SHUF_INS6]], i8 [[SHUF_ELT8]], i64 3
+// CHECK-NEXT:    [[SHUF_IDX10:%.*]] = extractelement <16 x i8> [[MASK]], i64 4
+// CHECK-NEXT:    [[SHUF_ELT11:%.*]] = extractelement <16 x i8> [[X]], i8 [[SHUF_IDX10]]
+// CHECK-NEXT:    [[SHUF_INS12:%.*]] = insertelement <16 x i8> [[SHUF_INS9]], i8 [[SHUF_ELT11]], i64 4
+// CHECK-NEXT:    [[SHUF_IDX13:%.*]] = extractelement <16 x i8> [[MASK]], i64 5
+// CHECK-NEXT:    [[SHUF_ELT14:%.*]] = extractelement <16 x i8> [[X]], i8 [[SHUF_IDX13]]
+// CHECK-NEXT:    [[SHUF_INS15:%.*]] = insertelement <16 x i8> [[SHUF_INS12]], i8 [[SHUF_ELT14]], i64 5
+// CHECK-NEXT:    [[SHUF_IDX16:%.*]] = extractelement <16 x i8> [[MASK]], i64 6
+// CHECK-NEXT:    [[SHUF_ELT17:%.*]] = extractelement <16 x i8> [[X]], i8 [[SHUF_IDX16]]
+// CHECK-NEXT:    [[SHUF_INS18:%.*]] = insertelement <16 x i8> [[SHUF_INS15]], i8 [[SHUF_ELT17]], i64 6
+// CHECK-NEXT:    [[SHUF_IDX19:%.*]] = extractelement <16 x i8> [[MASK]], i64 7
+// CHECK-NEXT:    [[SHUF_ELT20:%.*]] = extractelement <16 x i8> [[X]], i8 [[SHUF_IDX19]]
+// CHECK-NEXT:    [[SHUF_INS21:%.*]] = insertelement <16 x i8> [[SHUF_INS18]], i8 [[SHUF_ELT20]], i64 7
+// CHECK-NEXT:    [[SHUF_IDX22:%.*]] = extractelement <16 x i8> [[MASK]], i64 8
+// CHECK-NEXT:    [[SHUF_ELT23:%.*]] = extractelement <16 x i8> [[X]], i8 [[SHUF_IDX22]]
+// CHECK-NEXT:    [[SHUF_INS24:%.*]] = insertelement <16 x i8> [[SHUF_INS21]], i8 [[SHUF_ELT23]], i64 8
+// CHECK-NEXT:    [[SHUF_IDX25:%.*]] = extractelement <16 x i8> [[MASK]], i64 9
+// CHECK-NEXT:    [[SHUF_ELT26:%.*]] = extractelement <16 x i8> [[X]], i8 [[SHUF_IDX25]]
+// CHECK-NEXT:    [[SHUF_INS27:%.*]] = insertelement <16 x i8> [[SHUF_INS24]], i8 [[SHUF_ELT26]], i64 9
+// CHECK-NEXT:    [[SHUF_IDX28:%.*]] = extractelement <16 x i8> [[MASK]], i64 10
+// CHECK-NEXT:    [[SHUF_ELT29:%.*]] = extractelement <16 x i8> [[X]], i8 [[SHUF_IDX28]]
+// CHECK-NEXT:    [[SHUF_INS30:%.*]] = insertelement <16 x i8> [[SHUF_INS27]], i8 [[SHUF_ELT29]], i64 10
+// CHECK-NEXT:    [[SHUF_IDX31:%.*]] = extractelement <16 x i8> [[MASK]], i64 11
+// CHECK-NEXT:    [[SHUF_ELT32:%.*]] = extractelement <16 x i8> [[X]], i8 [[SHUF_IDX31]]
+// CHECK-NEXT:    [[SHUF_INS33:%.*]] = insertelement <16 x i8> [[SHUF_INS30]], i8 [[SHUF_ELT32]], i64 11
+// CHECK-NEXT:    [[SHUF_IDX34:%.*]] = extractelement <16 x i8> [[MASK]], i64 12
+// CHECK-NEXT:    [[SHUF_ELT35:%.*]] = extractelement <16 x i8> [[X]], i8 [[SHUF_IDX34]]
+// CHECK-NEXT:    [[SHUF_INS36:%.*]] = insertelement <16 x i8> [[SHUF_INS33]], i8 [[SHUF_ELT35]], i64 12
+// CHECK-NEXT:    [[SHUF_IDX37:%.*]] = extractelement <16 x i8> [[MASK]], i64 13
+// CHECK-NEXT:    [[SHUF_ELT38:%.*]] = extractelement <16 x i8> [[X]], i8 [[SHUF_IDX37]]
+// CHECK-NEXT:    [[SHUF_INS39:%.*]] = insertelement <16 x i8> [[SHUF_INS36]], i8 [[SHUF_ELT38]], i64 13
+// CHECK-NEXT:    [[SHUF_IDX40:%.*]] = extractelement <16 x i8> [[MASK]], i64 14
+// CHECK-NEXT:    [[SHUF_ELT41:%.*]] = extractelement <16 x i8> [[X]], i8 [[SHUF_IDX40]]
+// CHECK-NEXT:    [[SHUF_INS42:%.*]] = insertelement <16 x i8> [[SHUF_INS39]], i8 [[SHUF_ELT41]], i64 14
+// CHECK-NEXT:    [[SHUF_IDX43:%.*]] = extractelement <16 x i8> [[MASK]], i64 15
+// CHECK-NEXT:    [[SHUF_ELT44:%.*]] = extractelement <16 x i8> [[X]], i8 [[SHUF_IDX43]]
+// CHECK-NEXT:    [[SHUF_INS45:%.*]] = insertelement <16 x i8> [[SHUF_INS42]], i8 [[SHUF_ELT44]], i64 15
+// CHECK-NEXT:    ret <16 x i8> [[SHUF_INS45]]
+//
+mfloat8x16_t f4(mfloat8x16_t x, int8x16_t p) {
----------------
CarolineConcatto wrote:

nit: s/f4/f3/

https://github.com/llvm/llvm-project/pull/119031


More information about the llvm-commits mailing list