[llvm] 938cc57 - [AArch64][GlobalISel] Select G_ADD_LOW into a MOVaddr pseudo.

[Amara Emerson via llvm-commits]

Author: Amara Emerson
Date: 2020-06-09T16:47:58-07:00
New Revision: 938cc573ee1ae7fd628da0c2a9a0bcbc36c451e7

URL: https://github.com/llvm/llvm-project/commit/938cc573ee1ae7fd628da0c2a9a0bcbc36c451e7
DIFF: https://github.com/llvm/llvm-project/commit/938cc573ee1ae7fd628da0c2a9a0bcbc36c451e7.diff

LOG: [AArch64][GlobalISel] Select G_ADD_LOW into a MOVaddr pseudo.

This ensures that we match SelectionDAG behaviour by waiting until the expand
pseudos pass to generate ADRP + ADD pairs. Doing this at selection time for the
G_ADD_LOW is fine because by the time we get to selecting the G_ADD_LOW,
previous attempts to fold it into loads/stores must have failed.

Differential Revision: https://reviews.llvm.org/D81512

Added: 
    

Modified: 
    llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp
    llvm/test/CodeGen/AArch64/GlobalISel/call-translator-variadic-musttail.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp b/llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp
index e8137a8a5c1a..49a791414fbb 100644
--- a/llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp
+++ b/llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp
@@ -1919,9 +1919,28 @@ bool AArch64InstructionSelector::select(MachineInstr &I) {
     return selectBrJT(I, MRI);
 
   case AArch64::G_ADD_LOW: {
-    I.setDesc(TII.get(AArch64::ADDXri));
-    I.addOperand(MachineOperand::CreateImm(0));
-    return constrainSelectedInstRegOperands(I, TII, TRI, RBI);
+    // This op may have been separated from it's ADRP companion by the localizer
+    // or some other code motion pass. Given that many CPUs will try to
+    // macro fuse these operations anyway, select this into a MOVaddr pseudo
+    // which will later be expanded into an ADRP+ADD pair after scheduling.
+    MachineInstr *BaseMI = MRI.getVRegDef(I.getOperand(1).getReg());
+    if (BaseMI->getOpcode() != AArch64::ADRP) {
+      I.setDesc(TII.get(AArch64::ADDXri));
+      I.addOperand(MachineOperand::CreateImm(0));
+      return constrainSelectedInstRegOperands(I, TII, TRI, RBI);
+    }
+    assert(TM.getCodeModel() == CodeModel::Small &&
+           "Expected small code model");
+    MachineIRBuilder MIB(I);
+    auto Op1 = BaseMI->getOperand(1);
+    auto Op2 = I.getOperand(2);
+    auto MovAddr = MIB.buildInstr(AArch64::MOVaddr, {I.getOperand(0)}, {})
+                       .addGlobalAddress(Op1.getGlobal(), Op1.getOffset(),
+                                         Op1.getTargetFlags())
+                       .addGlobalAddress(Op2.getGlobal(), Op2.getOffset(),
+                                         Op2.getTargetFlags());
+    I.eraseFromParent();
+    return constrainSelectedInstRegOperands(*MovAddr, TII, TRI, RBI);
   }
 
   case TargetOpcode::G_BSWAP: {

diff  --git a/llvm/test/CodeGen/AArch64/GlobalISel/call-translator-variadic-musttail.ll b/llvm/test/CodeGen/AArch64/GlobalISel/call-translator-variadic-musttail.ll
index ef569b2f055a..0a0f94aa0a34 100644
--- a/llvm/test/CodeGen/AArch64/GlobalISel/call-translator-variadic-musttail.ll
+++ b/llvm/test/CodeGen/AArch64/GlobalISel/call-translator-variadic-musttail.ll
@@ -50,10 +50,11 @@ define i32 @test_musttail_variadic_spill(i32 %arg0, ...) {
 ; CHECK-NEXT:    .cfi_offset w26, -80
 ; CHECK-NEXT:    .cfi_offset w27, -88
 ; CHECK-NEXT:    .cfi_offset w28, -96
-; CHECK-NEXT:    mov x27, x8
-; CHECK-NEXT:    adrp x8, _asdf at PAGE
 ; CHECK-NEXT:    mov w19, w0
-; CHECK-NEXT:    add x0, x8, _asdf at PAGEOFF
+; CHECK-NEXT:  Lloh0:
+; CHECK-NEXT:    adrp x0, _asdf at PAGE
+; CHECK-NEXT:  Lloh1:
+; CHECK-NEXT:    add x0, x0, _asdf at PAGEOFF
 ; CHECK-NEXT:    mov x20, x1
 ; CHECK-NEXT:    mov x21, x2
 ; CHECK-NEXT:    mov x22, x3
@@ -65,6 +66,7 @@ define i32 @test_musttail_variadic_spill(i32 %arg0, ...) {
 ; CHECK-NEXT:    stp q3, q2, [sp, #64] ; 32-byte Folded Spill
 ; CHECK-NEXT:    stp q5, q4, [sp, #32] ; 32-byte Folded Spill
 ; CHECK-NEXT:    stp q7, q6, [sp] ; 32-byte Folded Spill
+; CHECK-NEXT:    mov x27, x8
 ; CHECK-NEXT:    bl _puts
 ; CHECK-NEXT:    ldp q1, q0, [sp, #96] ; 32-byte Folded Reload
 ; CHECK-NEXT:    ldp q3, q2, [sp, #64] ; 32-byte Folded Reload
@@ -87,6 +89,7 @@ define i32 @test_musttail_variadic_spill(i32 %arg0, ...) {
 ; CHECK-NEXT:    ldp x28, x27, [sp, #128] ; 16-byte Folded Reload
 ; CHECK-NEXT:    add sp, sp, #224 ; =224
 ; CHECK-NEXT:    b _musttail_variadic_callee
+; CHECK-NEXT:    .loh AdrpAdd Lloh0, Lloh1
   call void @puts(i8* getelementptr ([4 x i8], [4 x i8]* @asdf, i32 0, i32 0))
   %r = musttail call i32 (i32, ...) @musttail_variadic_callee(i32 %arg0, ...)
   ret i32 %r
@@ -189,16 +192,16 @@ define void @h_thunk(%struct.Foo* %this, ...) {
 ; CHECK-NEXT:    ldr x9, [x0, #8]
 ; CHECK-NEXT:    br x9
 ; CHECK-NEXT:  LBB5_2: ; %else
-; CHECK-NEXT:  Lloh0:
+; CHECK-NEXT:  Lloh2:
 ; CHECK-NEXT:    adrp x10, _g at GOTPAGE
 ; CHECK-NEXT:    ldr x9, [x0, #16]
-; CHECK-NEXT:  Lloh1:
+; CHECK-NEXT:  Lloh3:
 ; CHECK-NEXT:    ldr x10, [x10, _g at GOTPAGEOFF]
 ; CHECK-NEXT:    mov w11, #42
-; CHECK-NEXT:  Lloh2:
+; CHECK-NEXT:  Lloh4:
 ; CHECK-NEXT:    str w11, [x10]
 ; CHECK-NEXT:    br x9
-; CHECK-NEXT:    .loh AdrpLdrGotStr Lloh0, Lloh1, Lloh2
+; CHECK-NEXT:    .loh AdrpLdrGotStr Lloh2, Lloh3, Lloh4
   %cond_p = getelementptr %struct.Foo, %struct.Foo* %this, i32 0, i32 0
   %cond = load i1, i1* %cond_p
   br i1 %cond, label %then, label %else