Index: lib/Target/ARM/Thumb1InstrInfo.cpp =================================================================== --- lib/Target/ARM/Thumb1InstrInfo.cpp +++ lib/Target/ARM/Thumb1InstrInfo.cpp @@ -54,11 +54,17 @@ .addReg(SrcReg, getKillRegState(KillSrc)) .add(predOps(ARMCC::AL)); else { - // FIXME: The performance consequences of this are going to be atrocious. - // Some things to try that should be better: - // * 'mov hi, $src; mov $dst, hi', with hi as either r10 or r11 - // * 'movs $dst, $src' if cpsr isn't live - // See: http://lists.llvm.org/pipermail/llvm-dev/2014-August/075998.html + // FIXME: Can also use 'mov hi, $src; mov $dst, hi', + // with hi as either r10 or r11. + + const TargetRegisterInfo *RegInfo = st.getRegisterInfo(); + if (MBB.computeRegisterLiveness(RegInfo, ARM::CPSR, I) + == MachineBasicBlock::LQR_Dead) { + BuildMI(MBB, I, DL, get(ARM::tMOVSr), DestReg) + .addReg(SrcReg, getKillRegState(KillSrc)) + ->addRegisterDead(ARM::CPSR, RegInfo); + return; + } // 'MOV lo, lo' is unpredictable on < v6, so use the stack to do it BuildMI(MBB, I, DL, get(ARM::tPUSH)) Index: test/CodeGen/ARM/arm-and-tst-peephole.ll =================================================================== --- test/CodeGen/ARM/arm-and-tst-peephole.ll +++ test/CodeGen/ARM/arm-and-tst-peephole.ll @@ -150,16 +150,14 @@ ; ; THUMB-LABEL: test_tst_assessment: ; THUMB: @ BB#0: -; THUMB-NEXT: push {r0} -; THUMB-NEXT: pop {r2} +; THUMB-NEXT: movs r2, r0 ; THUMB-NEXT: movs r0, #1 ; THUMB-NEXT: ands r0, r2 ; THUMB-NEXT: subs r2, r0, #1 ; THUMB-NEXT: lsls r1, r1, #31 ; THUMB-NEXT: beq .LBB2_2 ; THUMB-NEXT: @ BB#1: -; THUMB-NEXT: push {r2} -; THUMB-NEXT: pop {r0} +; THUMB-NEXT: movs r0, r2 ; THUMB-NEXT: .LBB2_2: ; THUMB-NEXT: bx lr ; Index: test/CodeGen/ARM/atomic-cmpxchg.ll =================================================================== --- test/CodeGen/ARM/atomic-cmpxchg.ll +++ test/CodeGen/ARM/atomic-cmpxchg.ll @@ -24,14 +24,12 @@ ; CHECK-THUMB-LABEL: test_cmpxchg_res_i8 ; CHECK-THUMB: bl __sync_val_compare_and_swap_1 ; CHECK-THUMB-NOT: mov [[R1:r[0-7]]], r0 -; CHECK-THUMB: push {r0} -; CHECK-THUMB: pop {[[R1:r[0-7]]]} +; CHECK-THUMB: movs [[R1:r[0-7]]], r0 ; CHECK-THUMB: movs r0, #1 ; CHECK-THUMB: movs [[R2:r[0-9]+]], #0 ; CHECK-THUMB: cmp [[R1]], {{r[0-9]+}} ; CHECK-THUMB: beq -; CHECK-THUMB: push {[[R2]]} -; CHECK-THUMB: pop {r0} +; CHECK-THUMB: movs r0, [[R2]] ; CHECK-ARMV6-LABEL: test_cmpxchg_res_i8: ; CHECK-ARMV6-NEXT: .fnstart Index: test/CodeGen/Thumb/copy_thumb.ll =================================================================== --- test/CodeGen/Thumb/copy_thumb.ll +++ test/CodeGen/Thumb/copy_thumb.ll @@ -16,15 +16,9 @@ ; RUN: llc -mtriple=thumbv4t-none--eabi < %s | FileCheck %s --check-prefix=CHECK-NOLOLOMOV ; RUN: llc -mtriple=thumbv5-none--eabi < %s | FileCheck %s --check-prefix=CHECK-NOLOLOMOV ; CHECK-NOLOLOMOV-LABEL: foo -; CHECK-NOLOLOMOV-NOT: mov [[TMP:r[0-7]]], [[SRC1:r[01]]] -; CHECK-NOLOLOMOV: push {[[SRC1:r[01]]]} -; CHECK-NOLOLOMOV-NEXT: pop {[[TMP:r[0-7]]]} -; CHECK-NOLOLOMOV-NOT: mov [[TMP:r[0-7]]], [[SRC1:r[01]]] -; CHECK-NOLOLOMOV: push {[[SRC2:r[01]]]} -; CHECK-NOLOLOMOV-NEXT: pop {[[SRC1]]} -; CHECK-NOLOLOMOV-NOT: mov [[TMP:r[0-7]]], [[SRC1:r[01]]] -; CHECK-NOLOLOMOV: push {[[TMP]]} -; CHECK-NOLOLOMOV-NEXT: pop {[[SRC2]]} +; CHECK-NOLOLOMOV: movs [[TMP:r[0-7]]], [[SRC1:r[01]]] +; CHECK-NOLOLOMOV-NEXT: movs [[SRC1]], [[SRC2:r[01]]] +; CHECK-NOLOLOMOV-NEXT: movs [[SRC2]], [[TMP]] ; CHECK-NOLOLOMOV-LABEL: bar ; CHECK-NOLOLOMOV-LABEL: fnend Index: test/CodeGen/Thumb/thumb-shrink-wrapping.ll =================================================================== --- test/CodeGen/Thumb/thumb-shrink-wrapping.ll +++ test/CodeGen/Thumb/thumb-shrink-wrapping.ll @@ -505,14 +505,9 @@ ; CHECK-NEXT: str r1, {{\[}}[[TMP_SP]]] ; CHECK-NEXT: str r1, {{\[}}[[TMP_SP]], #4] ; CHECK-NEXT: str r1, {{\[}}[[TMP_SP]], #8] -; Thumb has quite a strange way for moving stuff -; in around. Oh well, match the current sequence. -; CHECK: push {r1} -; CHECK-NEXT: pop {r0} -; CHECK: push {r1} -; CHECK-NEXT: pop {r2} -; CHECK: push {r1} -; CHECK-NEXT: pop {r3} +; CHECK: movs r0, r1 +; CHECK-NEXT: movs r2, r1 +; CHECK-NEXT: movs r3, r1 ; CHECK-NEXT: bl ; CHECK-NEXT: lsls r0, r0, #3 ;