[llvm] [IA]: Construct (de)interleave4 out of (de)interleave2 (PR #89276)

[Paul Walker via llvm-commits]

================
@@ -0,0 +1,134 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 2
+; RUN: opt < %s -passes=interleaved-access -mtriple=aarch64-linux-gnu -mattr=+sve -S | FileCheck %s
+
+
+define void @deinterleave4(ptr %src) {
+; CHECK-LABEL: define void @deinterleave4
+; CHECK-SAME: (ptr [[SRC:%.*]]) #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT:    [[LDN:%.*]] = call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld4.sret.nxv4i32(<vscale x 4 x i1> shufflevector (<vscale x 4 x i1> insertelement (<vscale x 4 x i1> poison, i1 true, i64 0), <vscale x 4 x i1> poison, <vscale x 4 x i32> zeroinitializer), ptr [[SRC]])
+; CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[LDN]], 0
+; CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[LDN]], 1
+; CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[LDN]], 2
+; CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[LDN]], 3
+; CHECK-NEXT:    [[SUM:%.*]] = add <vscale x 4 x i32> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[SUB:%.*]] = sub <vscale x 4 x i32> [[TMP3]], [[TMP4]]
+; CHECK-NEXT:    ret void
+;
+
+  %load = load <vscale x 16 x i32>, ptr %src, align 4
+  %deinterleave_src = tail call { <vscale x 8 x i32>, <vscale x 8 x i32> } @llvm.vector.deinterleave2.nxv16i32(<vscale x 16 x i32> %load)
+  %3 = extractvalue { <vscale x 8 x i32>, <vscale x 8 x i32> } %deinterleave_src, 0
+  %4 = extractvalue { <vscale x 8 x i32>, <vscale x 8 x i32> } %deinterleave_src, 1
+  %deinterleave_half1 = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.vector.deinterleave2.nxv8i32(<vscale x 8 x i32> %3)
+  %5 = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } %deinterleave_half1, 0
+  %6 = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } %deinterleave_half1, 1
+  %deinterleave_half2 = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.vector.deinterleave2.nxv8i32(<vscale x 8 x i32> %4)
+  %7 = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } %deinterleave_half2, 0
+  %8 = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } %deinterleave_half2, 1
+  %sum = add <vscale x 4 x i32> %5, %7
+  %sub = sub <vscale x 4 x i32> %6, %8
+  ret void
+}
+
+define void @wide_deinterleave4(ptr %src) {
+; CHECK-LABEL: define void @wide_deinterleave4
+; CHECK-SAME: (ptr [[SRC:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 4 x i32>, ptr [[SRC]], i64 0
+; CHECK-NEXT:    [[LDN:%.*]] = call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld4.sret.nxv4i32(<vscale x 4 x i1> shufflevector (<vscale x 4 x i1> insertelement (<vscale x 4 x i1> poison, i1 true, i64 0), <vscale x 4 x i1> poison, <vscale x 4 x i32> zeroinitializer), ptr [[TMP1]])
+; CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[LDN]], 0
+; CHECK-NEXT:    [[TMP3:%.*]] = call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP2]], i64 0)
+; CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[LDN]], 1
+; CHECK-NEXT:    [[TMP5:%.*]] = call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP4]], i64 0)
+; CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[LDN]], 2
+; CHECK-NEXT:    [[TMP7:%.*]] = call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP6]], i64 0)
+; CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[LDN]], 3
+; CHECK-NEXT:    [[TMP9:%.*]] = call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP8]], i64 0)
+; CHECK-NEXT:    [[TMP10:%.*]] = getelementptr <vscale x 4 x i32>, ptr [[SRC]], i64 4
+; CHECK-NEXT:    [[LDN1:%.*]] = call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld4.sret.nxv4i32(<vscale x 4 x i1> shufflevector (<vscale x 4 x i1> insertelement (<vscale x 4 x i1> poison, i1 true, i64 0), <vscale x 4 x i1> poison, <vscale x 4 x i32> zeroinitializer), ptr [[TMP10]])
+; CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[LDN1]], 0
+; CHECK-NEXT:    [[TMP12:%.*]] = call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP11]], i64 4)
+; CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[LDN1]], 1
+; CHECK-NEXT:    [[TMP14:%.*]] = call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP13]], i64 4)
+; CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[LDN1]], 2
+; CHECK-NEXT:    [[TMP16:%.*]] = call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP7]], <vscale x 4 x i32> [[TMP15]], i64 4)
+; CHECK-NEXT:    [[TMP17:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[LDN1]], 3
+; CHECK-NEXT:    [[TMP18:%.*]] = call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP9]], <vscale x 4 x i32> [[TMP17]], i64 4)
+; CHECK-NEXT:    ret void
+;
+  %load = load <vscale x 32 x i32>, ptr %src, align 4
+  %deinterleave_src = tail call { <vscale x 16 x i32>, <vscale x 16 x i32> } @llvm.vector.deinterleave2.nxv32i32(<vscale x 32 x i32> %load)
+  %3 = extractvalue { <vscale x 16 x i32>, <vscale x 16 x i32> } %deinterleave_src, 0
+  %4 = extractvalue { <vscale x 16 x i32>, <vscale x 16 x i32> } %deinterleave_src, 1
+  %deinterleave_half1 = tail call { <vscale x 8 x i32>, <vscale x 8 x i32> } @llvm.vector.deinterleave2.nxv16i32(<vscale x 16 x i32> %3)
+  %5 = extractvalue { <vscale x 8 x i32>, <vscale x 8 x i32> } %deinterleave_half1, 0
+  %6 = extractvalue { <vscale x 8 x i32>, <vscale x 8 x i32> } %deinterleave_half1, 1
+  %deinterleave_half2 = tail call { <vscale x 8 x i32>, <vscale x 8 x i32> } @llvm.vector.deinterleave2.nxv16i32(<vscale x 16 x i32> %4)
+  %7 = extractvalue { <vscale x 8 x i32>, <vscale x 8 x i32> } %deinterleave_half2, 0
+  %8 = extractvalue { <vscale x 8 x i32>, <vscale x 8 x i32> } %deinterleave_half2, 1
+  ret void
----------------
paulwalker-arm wrote:

The other tests are fine but for `wide_deinterleave4` it is worth adding the add/sub combination that you have for `deinterleave4` because it better shows the extracted values end up where they're suppose to be.

https://github.com/llvm/llvm-project/pull/89276