[llvm] 22f423a - [ARM] Add some extra testing for MVE postinc loops. NFC
David Green via llvm-commits
llvm-commits at lists.llvm.org
Thu Sep 21 23:08:54 PDT 2023
Author: David Green
Date: 2023-09-22T07:08:49+01:00
New Revision: 22f423aa465b260112ff64a771408958e4d898ff
URL: https://github.com/llvm/llvm-project/commit/22f423aa465b260112ff64a771408958e4d898ff
DIFF: https://github.com/llvm/llvm-project/commit/22f423aa465b260112ff64a771408958e4d898ff.diff
LOG: [ARM] Add some extra testing for MVE postinc loops. NFC
Added:
llvm/test/CodeGen/Thumb2/mve-useafterloop.ll
Modified:
Removed:
################################################################################
diff --git a/llvm/test/CodeGen/Thumb2/mve-useafterloop.ll b/llvm/test/CodeGen/Thumb2/mve-useafterloop.ll
new file mode 100644
index 000000000000000..5f2d356a6d11951
--- /dev/null
+++ b/llvm/test/CodeGen/Thumb2/mve-useafterloop.ll
@@ -0,0 +1,142 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 3
+; RUN: llc -mtriple=thumbv8.1m.main-none-none-eabi -mattr=+mve.fp %s -o - | FileCheck %s
+
+define nonnull ptr @useafterloop(ptr nocapture noundef readonly %pSrcA, ptr nocapture noundef readonly %pSrcB, ptr noundef writeonly %pDst, i32 noundef %blockSize) {
+; CHECK-LABEL: useafterloop:
+; CHECK: @ %bb.0: @ %entry
+; CHECK-NEXT: .save {r7, lr}
+; CHECK-NEXT: push {r7, lr}
+; CHECK-NEXT: mov.w lr, #64
+; CHECK-NEXT: mov r12, r2
+; CHECK-NEXT: movs r3, #0
+; CHECK-NEXT: .LBB0_1: @ %while.body
+; CHECK-NEXT: @ =>This Inner Loop Header: Depth=1
+; CHECK-NEXT: vldrw.u32 q0, [r0], #16
+; CHECK-NEXT: vldrw.u32 q1, [r1], #16
+; CHECK-NEXT: add.w r2, r12, r3
+; CHECK-NEXT: adds r3, #16
+; CHECK-NEXT: vadd.f32 q0, q1, q0
+; CHECK-NEXT: vstrw.32 q0, [r2]
+; CHECK-NEXT: le lr, .LBB0_1
+; CHECK-NEXT: @ %bb.2: @ %while.end
+; CHECK-NEXT: mov r0, r12
+; CHECK-NEXT: pop {r7, pc}
+entry:
+ br label %while.body
+
+while.body:
+ %pSrcA.addr.012 = phi ptr [ %pSrcA, %entry ], [ %add.ptr, %while.body ]
+ %pSrcB.addr.011 = phi ptr [ %pSrcB, %entry ], [ %add.ptr1, %while.body ]
+ %pDst.addr.010 = phi ptr [ %pDst, %entry ], [ %add.ptr2, %while.body ]
+ %blkCnt.09 = phi i32 [ 64, %entry ], [ %dec, %while.body ]
+ %0 = load <4 x float>, ptr %pSrcA.addr.012, align 4
+ %1 = load <4 x float>, ptr %pSrcB.addr.011, align 4
+ %2 = fadd fast <4 x float> %1, %0
+ store <4 x float> %2, ptr %pDst.addr.010, align 4
+ %add.ptr = getelementptr inbounds float, ptr %pSrcA.addr.012, i32 4
+ %add.ptr1 = getelementptr inbounds float, ptr %pSrcB.addr.011, i32 4
+ %add.ptr2 = getelementptr inbounds float, ptr %pDst.addr.010, i32 4
+ %dec = add nsw i32 %blkCnt.09, -1
+ %cmp.not = icmp eq i32 %dec, 0
+ br i1 %cmp.not, label %while.end, label %while.body
+
+while.end:
+ ret ptr %pDst
+}
+
+
+define nonnull ptr @nouse(ptr nocapture noundef readonly %pSrcA, ptr nocapture noundef readonly %pSrcB, ptr noundef writeonly %pDst, i32 noundef %blockSize) {
+; CHECK-LABEL: nouse:
+; CHECK: @ %bb.0: @ %entry
+; CHECK-NEXT: .save {r7, lr}
+; CHECK-NEXT: push {r7, lr}
+; CHECK-NEXT: mov.w lr, #64
+; CHECK-NEXT: mov r3, r2
+; CHECK-NEXT: .LBB1_1: @ %while.body
+; CHECK-NEXT: @ =>This Inner Loop Header: Depth=1
+; CHECK-NEXT: vldrw.u32 q0, [r0], #16
+; CHECK-NEXT: vldrw.u32 q1, [r1], #16
+; CHECK-NEXT: vadd.f32 q0, q1, q0
+; CHECK-NEXT: vstrb.8 q0, [r3], #16
+; CHECK-NEXT: le lr, .LBB1_1
+; CHECK-NEXT: @ %bb.2: @ %while.end
+; CHECK-NEXT: adds r0, r2, #4
+; CHECK-NEXT: pop {r7, pc}
+entry:
+ br label %while.body
+
+while.body:
+ %pSrcA.addr.012 = phi ptr [ %pSrcA, %entry ], [ %add.ptr, %while.body ]
+ %pSrcB.addr.011 = phi ptr [ %pSrcB, %entry ], [ %add.ptr1, %while.body ]
+ %pDst.addr.010 = phi ptr [ %pDst, %entry ], [ %add.ptr2, %while.body ]
+ %blkCnt.09 = phi i32 [ 64, %entry ], [ %dec, %while.body ]
+ %0 = load <4 x float>, ptr %pSrcA.addr.012, align 4
+ %1 = load <4 x float>, ptr %pSrcB.addr.011, align 4
+ %2 = fadd fast <4 x float> %1, %0
+ store <4 x float> %2, ptr %pDst.addr.010, align 4
+ %add.ptr = getelementptr inbounds float, ptr %pSrcA.addr.012, i32 4
+ %add.ptr1 = getelementptr inbounds float, ptr %pSrcB.addr.011, i32 4
+ %add.ptr2 = getelementptr inbounds float, ptr %pDst.addr.010, i32 4
+ %dec = add nsw i32 %blkCnt.09, -1
+ %cmp.not = icmp eq i32 %dec, 0
+ br i1 %cmp.not, label %while.end, label %while.body
+
+while.end:
+ %add.ptr3 = getelementptr inbounds float, ptr %pDst, i32 1
+ ret ptr %add.ptr3
+}
+
+define nofpclass(nan inf) float @manyusesafterloop(ptr nocapture noundef readonly %pSrcA, ptr nocapture noundef readonly %pSrcB, ptr nocapture noundef %pDst, i32 noundef %blockSize) {
+; CHECK-LABEL: manyusesafterloop:
+; CHECK: @ %bb.0: @ %entry
+; CHECK-NEXT: .save {r4, lr}
+; CHECK-NEXT: push {r4, lr}
+; CHECK-NEXT: mov.w lr, #64
+; CHECK-NEXT: movs r3, #0
+; CHECK-NEXT: .LBB2_1: @ %while.body
+; CHECK-NEXT: @ =>This Inner Loop Header: Depth=1
+; CHECK-NEXT: add.w r12, r0, r3
+; CHECK-NEXT: adds r4, r1, r3
+; CHECK-NEXT: vldrw.u32 q1, [r4]
+; CHECK-NEXT: vldrw.u32 q0, [r12]
+; CHECK-NEXT: adds r4, r2, r3
+; CHECK-NEXT: adds r3, #16
+; CHECK-NEXT: vadd.f32 q0, q1, q0
+; CHECK-NEXT: vstrw.32 q0, [r4]
+; CHECK-NEXT: le lr, .LBB2_1
+; CHECK-NEXT: @ %bb.2: @ %while.end
+; CHECK-NEXT: vldr s0, [r2]
+; CHECK-NEXT: vldr s2, [r0]
+; CHECK-NEXT: vadd.f32 s0, s2, s0
+; CHECK-NEXT: vldr s2, [r1]
+; CHECK-NEXT: vadd.f32 s0, s0, s2
+; CHECK-NEXT: vmov r0, s0
+; CHECK-NEXT: pop {r4, pc}
+entry:
+ br label %while.body
+
+while.body:
+ %pSrcA.addr.016 = phi ptr [ %pSrcA, %entry ], [ %add.ptr, %while.body ]
+ %pSrcB.addr.015 = phi ptr [ %pSrcB, %entry ], [ %add.ptr1, %while.body ]
+ %pDst.addr.014 = phi ptr [ %pDst, %entry ], [ %add.ptr2, %while.body ]
+ %blkCnt.013 = phi i32 [ 64, %entry ], [ %dec, %while.body ]
+ %0 = load <4 x float>, ptr %pSrcA.addr.016, align 4
+ %1 = load <4 x float>, ptr %pSrcB.addr.015, align 4
+ %2 = fadd fast <4 x float> %1, %0
+ store <4 x float> %2, ptr %pDst.addr.014, align 4
+ %add.ptr = getelementptr inbounds float, ptr %pSrcA.addr.016, i32 4
+ %add.ptr1 = getelementptr inbounds float, ptr %pSrcB.addr.015, i32 4
+ %add.ptr2 = getelementptr inbounds float, ptr %pDst.addr.014, i32 4
+ %dec = add nsw i32 %blkCnt.013, -1
+ %cmp.not = icmp eq i32 %dec, 0
+ br i1 %cmp.not, label %while.end, label %while.body
+
+while.end:
+ %3 = load float, ptr %pDst, align 4
+ %4 = load float, ptr %pSrcA, align 4
+ %add = fadd fast float %4, %3
+ %5 = load float, ptr %pSrcB, align 4
+ %add5 = fadd fast float %add, %5
+ ret float %add5
+}
+
More information about the llvm-commits
mailing list