[ARM] Swap cmp operands for automatic shifts

Swap the compare operands if the lhs is a shift and the rhs isn't,
as in arm and T2 the shift can be performed by the compare for its
second operand.

Differential Revision: https://reviews.llvm.org/D39004

llvm-svn: 316562

3 files changed, 160 insertions, 52 deletions

diff --git a/llvm/lib/Target/ARM/ARMISelLowering.cpp b/llvm/lib/Target/ARM/ARMISelLowering.cpp
index 3f126024df3..7c8f6d99568 100644
--- a/llvm/lib/Target/ARM/ARMISelLowering.cpp
+++ b/llvm/lib/Target/ARM/ARMISelLowering.cpp
@@ -3857,6 +3857,12 @@ SDValue ARMTargetLowering::getARMCmp(SDValue LHS, SDValue RHS, ISD::CondCode CC,
         break;
       }
     }
+  } else if ((ARM_AM::getShiftOpcForNode(LHS.getOpcode()) != ARM_AM::no_shift) &&
+             (ARM_AM::getShiftOpcForNode(RHS.getOpcode()) == ARM_AM::no_shift)) {
+    // In ARM and Thumb-2, the compare instructions can shift their second
+    // operand.
+    CC = ISD::getSetCCSwappedOperands(CC);
+    std::swap(LHS, RHS);
   }
 
   ARMCC::CondCodes CondCode = IntCCToARMCC(CC);
diff --git a/llvm/test/CodeGen/ARM/cmp.ll b/llvm/test/CodeGen/ARM/cmp.ll
new file mode 100644
index 00000000000..e12a096b8c0
--- /dev/null
+++ b/llvm/test/CodeGen/ARM/cmp.ll
@@ -0,0 +1,154 @@
+; RUN: llc -mtriple=armv7 %s -o - | FileCheck %s
+; RUN: llc -mtriple=thumb-eabi -mcpu=arm1156t2-s -mattr=+thumb2 %s -o - | FileCheck %s --check-prefix=CHECK-T2
+
+define i1 @f1(i32 %a, i32 %b) {
+; CHECK-LABEL: f1:
+; CHECK: mov r2, #0
+; CHECK: cmp r0, r1
+; CHECK: movwne r2, #1
+; CHECK: mov r0, r2
+; CHECK-T2: mov{{.*}} r2, #0
+; CHECK-T2: cmp r0, r1
+; CHECK-T2: movne r2, #1
+; CHECK-T2: mov r0, r2
+    %tmp = icmp ne i32 %a, %b
+    ret i1 %tmp
+}
+
+define i1 @f2(i32 %a, i32 %b) {
+; CHECK-LABEL: f2:
+; CHECK: mov r2, #0
+; CHECK: cmp r0, r1
+; CHECK: movweq r2, #1
+; CHECK: mov r0, r2
+; CHECK-T2: mov{{.*}} r2, #0
+; CHECK-T2: cmp r0, r1
+; CHECK-T2: moveq r2, #1
+; CHECK-T2: mov r0, r2
+    %tmp = icmp eq i32 %a, %b
+    ret i1 %tmp
+}
+
+define i1 @f6(i32 %a, i32 %b) {
+; CHECK-LABEL: f6:
+; CHECK: mov r2, #0
+; CHECK: cmp {{.*}}, r1, lsl #5
+; CHECK: movweq r2, #1
+; CHECK: mov r0, r2
+; CHECK-T2: mov{{.*}} r2, #0
+; CHECK-T2: cmp.w r0, r1, lsl #5
+; CHECK-T2: moveq r2, #1
+; CHECK-T2: mov r0, r2
+    %tmp = shl i32 %b, 5
+    %tmp1 = icmp eq i32 %a, %tmp
+    ret i1 %tmp1
+}
+
+define i1 @f7(i32 %a, i32 %b) {
+; CHECK-LABEL: f7:
+; CHECK: mov r2, #0
+; CHECK: cmp r0, r1, lsr #6
+; CHECK: movwne r2, #1
+; CHECK: mov r0, r2
+; CHECK-T2: mov{{.*}} r2, #0
+; CHECK-T2: cmp.w r0, r1, lsr #6
+; CHECK-T2: movne r2, #1
+; CHECK-T2: mov r0, r2
+    %tmp = lshr i32 %b, 6
+    %tmp1 = icmp ne i32 %a, %tmp
+    ret i1 %tmp1
+}
+
+define i1 @f8(i32 %a, i32 %b) {
+; CHECK-LABEL: f8:
+; CHECK: mov r2, #0
+; CHECK: cmp r0, r1, asr #7
+; CHECK: movweq r2, #1
+; CHECK: mov r0, r2
+; CHECK-T2: mov{{.*}} r2, #0
+; CHECK-T2: cmp.w r0, r1, asr #7
+; CHECK-T2: moveq r2, #1
+; CHECK-T2: mov r0, r2
+    %tmp = ashr i32 %b, 7
+    %tmp1 = icmp eq i32 %a, %tmp
+    ret i1 %tmp1
+}
+
+define i1 @f9(i32 %a) {
+; CHECK-LABEL: f9:
+; CHECK: mov r1, #0
+; CHECK: cmp r0, r0, ror #8
+; CHECK: movwne r1, #1
+; CHECK: mov r0, r1
+; CHECK-T2: mov{{.*}} r1, #0
+; CHECK-T2: cmp.w r0, r0, ror #8
+; CHECK-T2: movne r1, #1
+; CHECK-T2: mov r0, r1
+    %l8 = shl i32 %a, 24
+    %r8 = lshr i32 %a, 8
+    %tmp = or i32 %l8, %r8
+    %tmp1 = icmp ne i32 %a, %tmp
+    ret i1 %tmp1
+}
+
+; CHECK-LABEL: swap_cmp_shl
+; CHECK: mov r2, #0
+; CHECK: cmp r1, r0, lsl #11
+; CHECK: movwlt r2, #1
+; CHECK-T2: mov{{.*}} r2, #0
+; CHECK-T2: cmp.w r1, r0, lsl #11
+; CHECK-T2: movlt r2, #1
+define arm_aapcscc i32 @swap_cmp_shl(i32 %a, i32 %b) {
+entry:
+  %shift = shl i32 %a, 11
+  %cmp = icmp sgt i32 %shift, %b
+  %conv = zext i1 %cmp to i32
+  ret i32 %conv
+}
+
+; CHECK-LABEL: swap_cmp_lshr
+; CHECK: mov r2, #0
+; CHECK: cmp r1, r0, lsr #11
+; CHECK: movwhi r2, #1
+; CHECK-T2: mov{{.*}} r2, #0
+; CHECK-T2: cmp.w r1, r0, lsr #11
+; CHECK-T2: movhi r2, #1
+define arm_aapcscc i32 @swap_cmp_lshr(i32 %a, i32 %b) {
+entry:
+  %shift = lshr i32 %a, 11
+  %cmp = icmp ult i32 %shift, %b
+  %conv = zext i1 %cmp to i32
+  ret i32 %conv
+}
+
+; CHECK-LABEL: swap_cmp_ashr
+; CHECK: mov r2, #0
+; CHECK: cmp r1, r0, asr #11
+; CHECK: movwle r2, #1
+; CHECK-T2: mov{{.*}} r2, #0
+; CHECK-T2: cmp.w r1, r0, asr #11
+; CHECK-T2: movle r2, #1
+define arm_aapcscc i32 @swap_cmp_ashr(i32 %a, i32 %b) {
+entry:
+  %shift = ashr i32 %a, 11
+  %cmp = icmp sge i32 %shift, %b
+  %conv = zext i1 %cmp to i32
+  ret i32 %conv
+}
+
+; CHECK-LABEL: swap_cmp_rotr
+; CHECK: mov r2, #0
+; CHECK: cmp r1, r0, ror #11
+; CHECK: movwls r2, #1
+; CHECK-T2: mov{{.*}} r2, #0
+; CHECK-T2: cmp.w r1, r0, ror #11
+; CHECK-T2: movls r2, #1
+define arm_aapcscc i32 @swap_cmp_rotr(i32 %a, i32 %b) {
+entry:
+  %lsr = lshr i32 %a, 11
+  %lsl = shl i32 %a, 21
+  %ror = or i32 %lsr, %lsl
+  %cmp = icmp uge i32 %ror, %b
+  %conv = zext i1 %cmp to i32
+  ret i32 %conv
+}
diff --git a/llvm/test/CodeGen/Thumb2/thumb2-cmp2.ll b/llvm/test/CodeGen/Thumb2/thumb2-cmp2.ll
deleted file mode 100644
index 4d840030f82..00000000000
--- a/llvm/test/CodeGen/Thumb2/thumb2-cmp2.ll
+++ /dev/null
@@ -1,52 +0,0 @@
-; RUN: llc -mtriple=thumb-eabi -mcpu=arm1156t2-s -mattr=+thumb2 %s -o - | FileCheck %s
-
-; These tests would be improved by 'movs r0, #0' being rematerialized below the
-; test as 'mov.w r0, #0'.
-
-define i1 @f1(i32 %a, i32 %b) {
-; CHECK-LABEL: f1:
-; CHECK: cmp {{.*}}, r1
-    %tmp = icmp ne i32 %a, %b
-    ret i1 %tmp
-}
-
-define i1 @f2(i32 %a, i32 %b) {
-; CHECK-LABEL: f2:
-; CHECK: cmp {{.*}}, r1
-    %tmp = icmp eq i32 %a, %b
-    ret i1 %tmp
-}
-
-define i1 @f6(i32 %a, i32 %b) {
-; CHECK-LABEL: f6:
-; CHECK: cmp.w {{.*}}, r1, lsl #5
-    %tmp = shl i32 %b, 5
-    %tmp1 = icmp eq i32 %tmp, %a
-    ret i1 %tmp1
-}
-
-define i1 @f7(i32 %a, i32 %b) {
-; CHECK-LABEL: f7:
-; CHECK: cmp.w {{.*}}, r1, lsr #6
-    %tmp = lshr i32 %b, 6
-    %tmp1 = icmp ne i32 %tmp, %a
-    ret i1 %tmp1
-}
-
-define i1 @f8(i32 %a, i32 %b) {
-; CHECK-LABEL: f8:
-; CHECK: cmp.w {{.*}}, r1, asr #7
-    %tmp = ashr i32 %b, 7
-    %tmp1 = icmp eq i32 %a, %tmp
-    ret i1 %tmp1
-}
-
-define i1 @f9(i32 %a, i32 %b) {
-; CHECK-LABEL: f9:
-; CHECK: cmp.w {{.*}}, {{.*}}, ror #8
-    %l8 = shl i32 %a, 24
-    %r8 = lshr i32 %a, 8
-    %tmp = or i32 %l8, %r8
-    %tmp1 = icmp ne i32 %a, %tmp
-    ret i1 %tmp1
-}