[LSR] Regenerate test checks (NFC)

kokkos · Jul 12, 2023 · 7a78756 · 7a78756
1 parent 5041442
commit 7a78756
Show file tree

Hide file tree

Showing 5 changed files with 554 additions and 150 deletions.
diff --git a/llvm/test/Transforms/LoopStrengthReduce/ARM/ivchain-ARM.ll b/llvm/test/Transforms/LoopStrengthReduce/ARM/ivchain-ARM.ll
@@ -1,17 +1,38 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 3
 ; RUN: llc -opaque-pointers=0 -O3 -mtriple=thumb-eabi -mcpu=cortex-a9 %s -o - | FileCheck %s -check-prefix=A9
 
 ; @simple is the most basic chain of address induction variables. Chaining
 ; saves at least one register and avoids complex addressing and setup
 ; code.
 ;
-; A9: @simple
 ; no expensive address computation in the preheader
-; A9: lsl
-; A9-NOT: lsl
-; A9: %loop
 ; no complex address modes
-; A9-NOT: lsl
 define i32 @simple(i32* %a, i32* %b, i32 %x) nounwind {
+; A9-LABEL: simple:
+; A9:       @ %bb.0: @ %entry
+; A9-NEXT:    .save {r4, r5, r6, lr}
+; A9-NEXT:    push {r4, r5, r6, lr}
+; A9-NEXT:    mov r3, r0
+; A9-NEXT:    lsls r2, r2, #2
+; A9-NEXT:    movs r0, #0
+; A9-NEXT:  .LBB0_1: @ %loop
+; A9-NEXT:    @ =>This Inner Loop Header: Depth=1
+; A9-NEXT:    add.w lr, r3, r2
+; A9-NEXT:    ldr.w r12, [r3, r2]
+; A9-NEXT:    ldr r3, [r3]
+; A9-NEXT:    add.w r4, lr, r2
+; A9-NEXT:    ldr.w r6, [lr, r2]
+; A9-NEXT:    add r0, r3
+; A9-NEXT:    adds r3, r4, r2
+; A9-NEXT:    add r0, r12
+; A9-NEXT:    ldr r5, [r4, r2]
+; A9-NEXT:    add r0, r6
+; A9-NEXT:    add r3, r2
+; A9-NEXT:    add r0, r5
+; A9-NEXT:    cmp r3, r1
+; A9-NEXT:    bne .LBB0_1
+; A9-NEXT:  @ %bb.2: @ %exit
+; A9-NEXT:    pop {r4, r5, r6, pc}
 entry:
   br label %loop
 loop:
@@ -37,15 +58,34 @@ exit:
 
 ; @user is not currently chained because the IV is live across memory ops.
 ;
-; A9: @user
 ; stride multiples computed in the preheader
-; A9: lsl
-; A9: lsl
-; A9: %loop
 ; complex address modes
-; A9: lsl
-; A9: lsl
 define i32 @user(i32* %a, i32* %b, i32 %x) nounwind {
+; A9-LABEL: user:
+; A9:       @ %bb.0: @ %entry
+; A9-NEXT:    .save {r4, r5, r6, r7, lr}
+; A9-NEXT:    push {r4, r5, r6, r7, lr}
+; A9-NEXT:    add.w r3, r2, r2, lsl #1
+; A9-NEXT:    lsl.w r12, r2, #4
+; A9-NEXT:    lsl.w lr, r3, #2
+; A9-NEXT:    movs r3, #0
+; A9-NEXT:  .LBB1_1: @ %loop
+; A9-NEXT:    @ =>This Inner Loop Header: Depth=1
+; A9-NEXT:    ldr r4, [r0]
+; A9-NEXT:    ldr.w r5, [r0, r2, lsl #3]
+; A9-NEXT:    ldr.w r6, [r0, r2, lsl #2]
+; A9-NEXT:    add r3, r4
+; A9-NEXT:    ldr.w r7, [r0, lr]
+; A9-NEXT:    add r3, r6
+; A9-NEXT:    add r3, r5
+; A9-NEXT:    add r3, r7
+; A9-NEXT:    str r3, [r0]
+; A9-NEXT:    add r0, r12
+; A9-NEXT:    cmp r0, r1
+; A9-NEXT:    bne .LBB1_1
+; A9-NEXT:  @ %bb.2: @ %exit
+; A9-NEXT:    mov r0, r3
+; A9-NEXT:    pop {r4, r5, r6, r7, pc}
 entry:
   br label %loop
 loop:
@@ -75,16 +115,43 @@ exit:
 ; used to do, and exactly what we don't want to do. LSR's new IV
 ; chaining feature should now undo the damage.
 ;
-; A9: extrastride:
 ; no spills
-; A9-NOT: str
 ; only one stride multiple in the preheader
-; A9: lsl
-; A9-NOT: {{str r|lsl}}
-; A9: %for.body{{$}}
 ; no complex address modes or reloads
-; A9-NOT: {{ldr .*[sp]|lsl}}
 define void @extrastride(i8* nocapture %main, i32 %main_stride, i32* nocapture %res, i32 %x, i32 %y, i32 %z) nounwind {
+; A9-LABEL: extrastride:
+; A9:       @ %bb.0: @ %entry
+; A9-NEXT:    .save {r4, r5, r6, r7, lr}
+; A9-NEXT:    push {r4, r5, r6, r7, lr}
+; A9-NEXT:    ldr.w r12, [sp, #24]
+; A9-NEXT:    cmp.w r12, #0
+; A9-NEXT:    beq .LBB2_3
+; A9-NEXT:  @ %bb.1: @ %for.body.lr.ph
+; A9-NEXT:    ldr r4, [sp, #20]
+; A9-NEXT:    add.w lr, r3, r1
+; A9-NEXT:    lsls r3, r4, #2
+; A9-NEXT:  .LBB2_2: @ %for.body
+; A9-NEXT:    @ =>This Inner Loop Header: Depth=1
+; A9-NEXT:    adds r5, r0, r1
+; A9-NEXT:    ldr r4, [r0, r1]
+; A9-NEXT:    ldr r0, [r0]
+; A9-NEXT:    subs.w r12, r12, #1
+; A9-NEXT:    ldr r6, [r5, r1]
+; A9-NEXT:    add r5, r1
+; A9-NEXT:    add r0, r4
+; A9-NEXT:    ldr r7, [r5, r1]
+; A9-NEXT:    add r5, r1
+; A9-NEXT:    add r0, r6
+; A9-NEXT:    ldr r4, [r5, r1]
+; A9-NEXT:    add r0, r7
+; A9-NEXT:    add r0, r4
+; A9-NEXT:    str r0, [r2]
+; A9-NEXT:    add.w r0, r5, r1
+; A9-NEXT:    add r2, r3
+; A9-NEXT:    add r0, lr
+; A9-NEXT:    bne .LBB2_2
+; A9-NEXT:  .LBB2_3: @ %for.end
+; A9-NEXT:    pop {r4, r5, r6, r7, pc}
 entry:
   %cmp8 = icmp eq i32 %z, 0
   br i1 %cmp8, label %for.end, label %for.body.lr.ph
@@ -136,10 +203,38 @@ for.end:                                          ; preds = %for.body, %entry
 ;  }
 ; where 's' can be folded into the addressing mode.
 ; Consequently, we should *not* form any chains.
-;
-; A9: foldedidx:
-; A9: ldrb{{(.w)?}} {{r[0-9]|lr}}, [{{r[0-9]|lr}}, #3]
 define void @foldedidx(i8* nocapture %a, i8* nocapture %b, i8* nocapture %c) nounwind ssp {
+; A9-LABEL: foldedidx:
+; A9:       @ %bb.0: @ %entry
+; A9-NEXT:    .save {r4, r5, r6, lr}
+; A9-NEXT:    push {r4, r5, r6, lr}
+; A9-NEXT:    mov.w lr, #0
+; A9-NEXT:  .LBB3_1: @ %for.body
+; A9-NEXT:    @ =>This Inner Loop Header: Depth=1
+; A9-NEXT:    ldrb.w r12, [r0, lr]
+; A9-NEXT:    add.w r4, r1, lr
+; A9-NEXT:    ldrb.w r3, [r1, lr]
+; A9-NEXT:    add r3, r12
+; A9-NEXT:    strb.w r3, [r2, lr]
+; A9-NEXT:    add.w r3, r0, lr
+; A9-NEXT:    ldrb.w r12, [r3, #1]
+; A9-NEXT:    ldrb r5, [r4, #1]
+; A9-NEXT:    add r12, r5
+; A9-NEXT:    add.w r5, r2, lr
+; A9-NEXT:    strb.w r12, [r5, #1]
+; A9-NEXT:    add.w lr, lr, #4
+; A9-NEXT:    cmp.w lr, #400
+; A9-NEXT:    ldrb.w r12, [r3, #2]
+; A9-NEXT:    ldrb r6, [r4, #2]
+; A9-NEXT:    add r6, r12
+; A9-NEXT:    strb r6, [r5, #2]
+; A9-NEXT:    ldrb r3, [r3, #3]
+; A9-NEXT:    ldrb r6, [r4, #3]
+; A9-NEXT:    add r3, r6
+; A9-NEXT:    strb r3, [r5, #3]
+; A9-NEXT:    bne .LBB3_1
+; A9-NEXT:  @ %bb.2: @ %for.end
+; A9-NEXT:    pop {r4, r5, r6, pc}
 entry:
   br label %for.body
 
@@ -200,14 +295,45 @@ for.end:                                          ; preds = %for.body
 ;
 ; Loads and stores should use post-increment addressing, no add's or add.w's.
 ; Most importantly, there should be no spills or reloads!
-;
-; A9: testNeon:
-; A9: %.lr.ph
-; A9-NOT: lsl.w
-; A9-NOT: {{ldr|str|adds|add r}}
-; A9-NOT: add.w r
-; A9: bne
 define hidden void @testNeon(i8* %ref_data, i32 %ref_stride, i32 %limit, <16 x i8>* nocapture %data) nounwind optsize {
+; A9-LABEL: testNeon:
+; A9:       @ %bb.0:
+; A9-NEXT:    .save {r4, r5, r7, lr}
+; A9-NEXT:    push {r4, r5, r7, lr}
+; A9-NEXT:    vmov.i32 q8, #0x0
+; A9-NEXT:    cmp r2, #1
+; A9-NEXT:    blt .LBB4_4
+; A9-NEXT:  @ %bb.1: @ %.lr.ph
+; A9-NEXT:    movs r5, #0
+; A9-NEXT:    movw r4, #64464
+; A9-NEXT:    sub.w r12, r5, r2, lsl #6
+; A9-NEXT:    sub.w lr, r1, r1, lsl #4
+; A9-NEXT:    movt r4, #65535
+; A9-NEXT:    mov r5, r3
+; A9-NEXT:  .LBB4_2: @ =>This Inner Loop Header: Depth=1
+; A9-NEXT:    vld1.64 {d18}, [r0], r1
+; A9-NEXT:    subs r2, #1
+; A9-NEXT:    vld1.64 {d19}, [r0], r1
+; A9-NEXT:    vst1.8 {d18, d19}, [r5]!
+; A9-NEXT:    vld1.64 {d20}, [r0], r1
+; A9-NEXT:    vld1.64 {d21}, [r0], r1
+; A9-NEXT:    vst1.8 {d20, d21}, [r5]!
+; A9-NEXT:    vld1.64 {d22}, [r0], r1
+; A9-NEXT:    vadd.i8 q9, q9, q10
+; A9-NEXT:    vld1.64 {d23}, [r0], r1
+; A9-NEXT:    vst1.8 {d22, d23}, [r5]!
+; A9-NEXT:    vld1.64 {d20}, [r0], r1
+; A9-NEXT:    vadd.i8 q9, q9, q11
+; A9-NEXT:    vld1.64 {d21}, [r0], lr
+; A9-NEXT:    vadd.i8 q9, q9, q10
+; A9-NEXT:    vadd.i8 q8, q8, q9
+; A9-NEXT:    vst1.8 {d20, d21}, [r5], r4
+; A9-NEXT:    bne .LBB4_2
+; A9-NEXT:  @ %bb.3: @ %._crit_edge
+; A9-NEXT:    add.w r3, r3, r12, lsl #4
+; A9-NEXT:  .LBB4_4:
+; A9-NEXT:    vst1.32 {d16, d17}, [r3]
+; A9-NEXT:    pop {r4, r5, r7, pc}
   %1 = icmp sgt i32 %limit, 0
   br i1 %1, label %.lr.ph, label %45
 
@@ -284,24 +410,41 @@ declare <1 x i64> @llvm.arm.neon.vld1.v1i64.p0i8(i8*, i32) nounwind readonly
 ; Handle chains in which the same offset is used for both loads and
 ; stores to the same array.
 ; rdar://11410078.
-;
-; A9: @testReuse
-; A9: %for.body
-; A9: vld1.8 {d{{[0-9]+}}}, [[BASE:[r[0-9]+]]], [[INC:r[0-9]]]
-; A9: vld1.8 {d{{[0-9]+}}}, [[BASE]], [[INC]]
-; A9: vld1.8 {d{{[0-9]+}}}, [[BASE]], [[INC]]
-; A9: vld1.8 {d{{[0-9]+}}}, [[BASE]], [[INC]]
-; A9: vld1.8 {d{{[0-9]+}}}, [[BASE]], [[INC]]
-; A9: vld1.8 {d{{[0-9]+}}}, [[BASE]], [[INC]]
-; A9: vld1.8 {d{{[0-9]+}}}, [[BASE]], [[INC]]
-; A9: vst1.8 {d{{[0-9]+}}}, [[BASE]], [[INC]]
-; A9: vst1.8 {d{{[0-9]+}}}, [[BASE]], [[INC]]
-; A9: vst1.8 {d{{[0-9]+}}}, [[BASE]], [[INC]]
-; A9: vst1.8 {d{{[0-9]+}}}, [[BASE]], [[INC]]
-; A9: vst1.8 {d{{[0-9]+}}}, [[BASE]], [[INC]]
-; A9: vst1.8 {d{{[0-9]+}}}, [[BASE]]
-; A9: bne
 define void @testReuse(i8* %src, i32 %stride) nounwind ssp {
+; A9-LABEL: testReuse:
+; A9:       @ %bb.0: @ %entry
+; A9-NEXT:    sub.w r12, r0, r1, lsl #2
+; A9-NEXT:    sub.w r0, r1, r1, lsl #2
+; A9-NEXT:    lsls r2, r0, #1
+; A9-NEXT:    movs r3, #0
+; A9-NEXT:  .LBB5_1: @ %for.body
+; A9-NEXT:    @ =>This Inner Loop Header: Depth=1
+; A9-NEXT:    add.w r0, r12, r3
+; A9-NEXT:    adds r3, #8
+; A9-NEXT:    vld1.8 {d16}, [r0], r1
+; A9-NEXT:    cmp r3, #32
+; A9-NEXT:    vld1.8 {d17}, [r0], r1
+; A9-NEXT:    vhadd.u8 d16, d16, d17
+; A9-NEXT:    vld1.8 {d18}, [r0], r1
+; A9-NEXT:    vhadd.u8 d17, d17, d18
+; A9-NEXT:    vld1.8 {d19}, [r0], r1
+; A9-NEXT:    vhadd.u8 d18, d18, d19
+; A9-NEXT:    vld1.8 {d20}, [r0], r1
+; A9-NEXT:    vhadd.u8 d19, d19, d20
+; A9-NEXT:    vld1.8 {d21}, [r0], r1
+; A9-NEXT:    vhadd.u8 d20, d20, d21
+; A9-NEXT:    vld1.8 {d22}, [r0], r1
+; A9-NEXT:    vhadd.u8 d21, d21, d22
+; A9-NEXT:    vld1.8 {d23}, [r0], r2
+; A9-NEXT:    vst1.8 {d16}, [r0], r1
+; A9-NEXT:    vst1.8 {d17}, [r0], r1
+; A9-NEXT:    vst1.8 {d18}, [r0], r1
+; A9-NEXT:    vst1.8 {d19}, [r0], r1
+; A9-NEXT:    vst1.8 {d20}, [r0], r1
+; A9-NEXT:    vst1.8 {d21}, [r0]
+; A9-NEXT:    bne .LBB5_1
+; A9-NEXT:  @ %bb.2: @ %for.end
+; A9-NEXT:    bx lr
 entry:
   %mul = shl nsw i32 %stride, 2
   %idx.neg = sub i32 0, %mul