[PowerPC] Support PIC Secure PLT for CALL_RM

https://reviews.llvm.org/D111433 introduced PPCISD::CALL_RM for
-frounding-math. -msecure-plt -frounding-math {-fpic,-fPIC} codegen for
PPC32 became incorrect when a function contains function calls but no
global variable references (GlobalBaseReg).

As reported by @q66 , musl/src/dirent/closedir.c implements such a
function, which is miscompiled.

PPCISD::CALL has custom logic to set up the base register
(https://reviews.llvm.org/D42112). Add an extra case for CALL_RM.

While here, improve the test to

* actually test `case PPCISD::CALL`: we need a non-leaf function that
  doesn't access global variables (global variables lead to
  GlobalBaseReg, which call `getGlobalBaseReg()` as well).
* test `ExternalSymbolSDNode` with a memset.

Supersedes: #72758

Pull Request: #121281

Author: MaskRay

llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp

@@ -5473,10 +5473,10 @@ void PPCDAGToDAGISel::Select(SDNode *N) {
     // generate secure plt code for TLS symbols.
     getGlobalBaseReg();
   } break;
-  case PPCISD::CALL: {
-    if (PPCLowering->getPointerTy(CurDAG->getDataLayout()) != MVT::i32 ||
-        !TM.isPositionIndependent() || !Subtarget->isSecurePlt() ||
-        !Subtarget->isTargetELF())
+  case PPCISD::CALL:
+  case PPCISD::CALL_RM: {
+    if (Subtarget->isPPC64() || !TM.isPositionIndependent() ||
+        !Subtarget->isSecurePlt() || !Subtarget->isTargetELF())
       break;
 
     SDValue Op = N->getOperand(1);
@@ -5489,8 +5489,7 @@ void PPCDAGToDAGISel::Select(SDNode *N) {
       if (ES->getTargetFlags() == PPCII::MO_PLT)
         getGlobalBaseReg();
     }
-  }
-    break;
+  } break;
 
   case PPCISD::GlobalBaseReg:
     ReplaceNode(N, getGlobalBaseReg());

llvm/test/CodeGen/PowerPC/ppc32-pic-large.ll

@@ -13,6 +13,8 @@ $bar1 = comdat any
 @bar2 = global i32 0, align 4, comdat($bar1)
 
 declare i32 @call_foo(i32, ...)
+declare i32 @call_strictfp() strictfp
+declare void @llvm.memset.p0.i64(ptr nocapture writeonly, i8, i64, i1 immarg)
 
 define i32 @foo() {
 entry:
@@ -21,6 +23,23 @@ entry:
   ret i32 %0
 }
 
+define i32 @foo1() strictfp {
+entry:
+  %call = call i32 (i32, ...) @call_foo(i32 0)
+  ret i32 %call
+}
+
+define i32 @foo1_strictfp() strictfp {
+entry:
+  %call = call i32 () @call_strictfp()
+  ret i32 %call
+}
+
+define void @foo2(ptr %a) {
+  call void @llvm.memset.p0.i64(ptr align 1 %a, i8 1, i64 1000, i1 false)
+  ret void
+}
+
 define i32 @load() {
 entry:
   %0 = load i32, ptr @bar1
@@ -49,6 +68,31 @@ entry:
 ; LARGE-SECUREPLT:   addi 30, 30, .LTOC-.L0$pb@l
 ; LARGE-SECUREPLT:   bl call_foo@PLT+32768
 
+; LARGE-SECUREPLT-LABEL: foo1:
+; LARGE-SECUREPLT:       .L1$pb:
+; LARGE-SECUREPLT-NEXT:    crxor 6, 6, 6
+; LARGE-SECUREPLT-NEXT:    mflr 30
+; LARGE-SECUREPLT-NEXT:    addis 30, 30, .LTOC-.L1$pb@ha
+; LARGE-SECUREPLT-NEXT:    addi 30, 30, .LTOC-.L1$pb@l
+; LARGE-SECUREPLT-NEXT:    li 3, 0
+; LARGE-SECUREPLT-NEXT:    bl call_foo@PLT+32768
+
+; LARGE-SECUREPLT-LABEL: foo1_strictfp:
+; LARGE-SECUREPLT:       .L2$pb:
+; LARGE-SECUREPLT-NEXT:    mflr 30
+; LARGE-SECUREPLT-NEXT:    addis 30, 30, .LTOC-.L2$pb@ha
+; LARGE-SECUREPLT-NEXT:    addi 30, 30, .LTOC-.L2$pb@l
+; LARGE-SECUREPLT-NEXT:    bl call_strictfp@PLT+32768
+
+; LARGE-SECUREPLT-LABEL: foo2:
+; LARGE-SECUREPLT:       .L3$pb:
+; LARGE-SECUREPLT:         mflr 30
+; LARGE-SECUREPLT-NEXT:    addis 30, 30, .LTOC-.L3$pb@ha
+; LARGE-SECUREPLT-NEXT:    addi 30, 30, .LTOC-.L3$pb@l
+; LARGE-SECUREPLT:         bl memset@PLT+32768
+
+; LARGE-SECUREPLT-LABEEL: load:
+
 ; LARGE:      .section .bss.bar1,"awG",@nobits,bar1,comdat
 ; LARGE:      bar1:
 ; LARGE:      .section .bss.bar2,"awG",@nobits,bar1,comdat