1 ; RUN: llc < %s -mtriple=thumbv7-apple-ios -mcpu=cortex-a8 -regalloc=fast | FileCheck %s -check-prefix=A8 2 ; RUN: llc < %s -mtriple=thumbv7-apple-ios -mcpu=cortex-m3 -regalloc=fast | FileCheck %s -check-prefix=M3 3 ; rdar://6949835 4 ; RUN: llc < %s -mtriple=thumbv7-apple-ios -mcpu=cortex-a8 -regalloc=basic | FileCheck %s -check-prefix=BASIC 5 ; RUN: llc < %s -mtriple=thumbv7-apple-ios -mcpu=cortex-a8 -regalloc=greedy | FileCheck %s -check-prefix=GREEDY 6 7 ; Magic ARM pair hints works best with linearscan / fast. 8 9 ; Cortex-M3 errata 602117: LDRD with base in list may result in incorrect base 10 ; register when interrupted or faulted. 11 12 @b = external global i64* 13 14 define i64 @t(i64 %a) nounwind readonly { 15 entry: 16 ; A8: t: 17 ; A8: ldrd r2, r3, [r2] 18 19 ; M3: t: 20 ; M3-NOT: ldrd 21 ; M3: ldm.w r2, {r2, r3} 22 23 %0 = load i64** @b, align 4 24 %1 = load i64* %0, align 4 25 %2 = mul i64 %1, %a 26 ret i64 %2 27 } 28 29 ; rdar://10435045 mixed LDRi8/LDRi12 30 ; 31 ; In this case, LSR generate a sequence of LDRi8/LDRi12. We should be 32 ; able to generate an LDRD pair here, but this is highly sensitive to 33 ; regalloc hinting. So, this doubles as a register allocation 34 ; test. RABasic currently does a better job within the inner loop 35 ; because of its *lack* of hinting ability. Whereas RAGreedy keeps 36 ; R0/R1/R2 live as the three arguments, forcing the LDRD's odd 37 ; destination into R3. We then sensibly split LDRD again rather then 38 ; evict another live range or use callee saved regs. Sorry if the test 39 ; is sensitive to Regalloc changes, but it is an interesting case. 40 ; 41 ; BASIC: @f 42 ; BASIC: %bb 43 ; BASIC: ldrd 44 ; BASIC: str 45 ; GREEDY: @f 46 ; GREEDY: %bb 47 ; GREEDY: ldrd 48 ; GREEDY: str 49 define void @f(i32* nocapture %a, i32* nocapture %b, i32 %n) nounwind { 50 entry: 51 %0 = add nsw i32 %n, -1 ; <i32> [#uses=2] 52 %1 = icmp sgt i32 %0, 0 ; <i1> [#uses=1] 53 br i1 %1, label %bb, label %return 54 55 bb: ; preds = %bb, %entry 56 %i.03 = phi i32 [ %tmp, %bb ], [ 0, %entry ] ; <i32> [#uses=3] 57 %scevgep = getelementptr i32* %a, i32 %i.03 ; <i32*> [#uses=1] 58 %scevgep4 = getelementptr i32* %b, i32 %i.03 ; <i32*> [#uses=1] 59 %tmp = add i32 %i.03, 1 ; <i32> [#uses=3] 60 %scevgep5 = getelementptr i32* %a, i32 %tmp ; <i32*> [#uses=1] 61 %2 = load i32* %scevgep, align 4 ; <i32> [#uses=1] 62 %3 = load i32* %scevgep5, align 4 ; <i32> [#uses=1] 63 %4 = add nsw i32 %3, %2 ; <i32> [#uses=1] 64 store i32 %4, i32* %scevgep4, align 4 65 %exitcond = icmp eq i32 %tmp, %0 ; <i1> [#uses=1] 66 br i1 %exitcond, label %return, label %bb 67 68 return: ; preds = %bb, %entry 69 ret void 70 } 71
