[PowerPC] Expand global named register support (#113482)

Enable all valid registers for intrinsics that read from and write
to global named registers.

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

@@ -17367,25 +17367,33 @@ SDValue PPCTargetLowering::LowerFRAMEADDR(SDValue Op,
   return FrameAddr;
 }
 
-// FIXME? Maybe this could be a TableGen attribute on some registers and
-// this table could be generated automatically from RegInfo.
-Register PPCTargetLowering::getRegisterByName(const char* RegName, LLT VT,
-                                              const MachineFunction &MF) const {
-  bool isPPC64 = Subtarget.isPPC64();
+#define GET_REGISTER_MATCHER
+#include "PPCGenAsmMatcher.inc"
 
-  bool is64Bit = isPPC64 && VT == LLT::scalar(64);
-  if (!is64Bit && VT != LLT::scalar(32))
+Register PPCTargetLowering::getRegisterByName(const char *RegName, LLT VT,
+                                              const MachineFunction &MF) const {
+  bool IsPPC64 = Subtarget.isPPC64();
+
+  bool Is64Bit = IsPPC64 && VT == LLT::scalar(64);
+  if (!Is64Bit && VT != LLT::scalar(32))
     report_fatal_error("Invalid register global variable type");
 
-  Register Reg = StringSwitch<Register>(RegName)
-                     .Case("r1", is64Bit ? PPC::X1 : PPC::R1)
-                     .Case("r2", isPPC64 ? Register() : PPC::R2)
-                     .Case("r13", (is64Bit ? PPC::X13 : PPC::R13))
-                     .Default(Register());
+  Register Reg = MatchRegisterName(RegName);
+  if (!Reg)
+    report_fatal_error(
+        Twine("Invalid global name register \"" + StringRef(RegName) + "\"."));
 
-  if (Reg)
-    return Reg;
-  report_fatal_error("Invalid register name global variable");
+  // FIXME: Unable to generate code for `-O2` but okay for `-O0`.
+  // Need followup investigation as to why.
+  if ((IsPPC64 && Reg == PPC::R2) || Reg == PPC::R0)
+    report_fatal_error(Twine("Trying to reserve an invalid register \"" +
+                             StringRef(RegName) + "\"."));
+
+  // Convert GPR to GP8R register for 64bit.
+  if (Is64Bit && StringRef(RegName).starts_with_insensitive("r"))
+    Reg = Reg.id() - PPC::R0 + PPC::X0;
+
+  return Reg;
 }
 
 bool PPCTargetLowering::isAccessedAsGotIndirect(SDValue GA) const {

llvm/test/CodeGen/PowerPC/named-reg-alloc-r0.ll

@@ -1,11 +1,9 @@
 ; RUN: not --crash llc < %s -mtriple=powerpc-unknown-linux-gnu 2>&1 | FileCheck %s
-; RUN: not --crash llc < %s -mtriple=powerpc-unknown-linux-gnu 2>&1 | FileCheck %s
 ; RUN: not --crash llc < %s -mtriple=powerpc64-unknown-linux-gnu 2>&1 | FileCheck %s
 
 define i32 @get_reg() nounwind {
 entry:
-; FIXME: Include an allocatable-specific error message
-; CHECK: Invalid register name global variable
+; CHECK: Trying to reserve an invalid register "r0".
         %reg = call i32 @llvm.read_register.i32(metadata !0)
   ret i32 %reg
 }

llvm/test/CodeGen/PowerPC/named-reg-alloc-r2-64.ll

@@ -1,10 +1,8 @@
 ; RUN: not --crash llc < %s -mtriple=powerpc64-unknown-linux-gnu 2>&1 | FileCheck %s
-; RUN: not --crash llc < %s -mtriple=powerpc64-unknown-linux-gnu 2>&1 | FileCheck %s
 
 define i64 @get_reg() nounwind {
 entry:
-; FIXME: Include an allocatable-specific error message
-; CHECK: Invalid register name global variable
+; CHECK: Trying to reserve an invalid register "r2".
         %reg = call i64 @llvm.read_register.i64(metadata !0)
   ret i64 %reg
 }

llvm/test/CodeGen/PowerPC/named-reg-alloc-r2.ll

@@ -3,11 +3,9 @@
 
 define i32 @get_reg() nounwind {
 entry:
-; FIXME: Include an allocatable-specific error message
-; CHECK-NOTPPC32: Invalid register name global variable
+; CHECK-NOTPPC32: Trying to reserve an invalid register "r2".
         %reg = call i32 @llvm.read_register.i32(metadata !0)
   ret i32 %reg
-
 ; CHECK-LABEL: @get_reg
 ; CHECK: mr 3, 2
 }

llvm/test/CodeGen/PowerPC/named-reg-alloc.ll

@@ -0,0 +1,144 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -O0 -verify-machineinstrs < %s -mtriple=powerpc-unknown-linux-gnu 2>&1 | FileCheck %s
+; RUN: llc -O0 -verify-machineinstrs < %s -mtriple=powerpc64-unknown-linux-gnu 2>&1 | FileCheck %s --check-prefix=CHECK64
+
+@mVal = dso_local global i32 15, align 4
+@myGVal = dso_local global i32 0, align 4
+
+define dso_local void @testSetIntReg(i32 noundef signext %xx) {
+; CHECK-LABEL: testSetIntReg:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    mr 5, 3
+; CHECK-NEXT:    blr
+;
+; CHECK64-LABEL: testSetIntReg:
+; CHECK64:       # %bb.0: # %entry
+; CHECK64-NEXT:    mr 5, 3
+; CHECK64-NEXT:    blr
+entry:
+  tail call void @llvm.write_register.i32(metadata !0, i32 %xx)
+  ret void
+}
+
+declare void @llvm.write_register.i32(metadata, i32)
+
+define dso_local signext range(i32 0, 2) i32 @testCmpReg() {
+; CHECK-LABEL: testCmpReg:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    lis 3, mVal@ha
+; CHECK-NEXT:    lwz 3, mVal@l(3)
+; CHECK-NEXT:    xori 3, 3, 15
+; CHECK-NEXT:    cntlzw 3, 3
+; CHECK-NEXT:    srwi 3, 3, 5
+; CHECK-NEXT:    blr
+;
+; CHECK64-LABEL: testCmpReg:
+; CHECK64:       # %bb.0: # %entry
+; CHECK64-NEXT:    addis 3, 2, mVal@toc@ha
+; CHECK64-NEXT:    addi 3, 3, mVal@toc@l
+; CHECK64-NEXT:    lwz 3, 0(3)
+; CHECK64-NEXT:    xori 3, 3, 15
+; CHECK64-NEXT:    cntlzw 3, 3
+; CHECK64-NEXT:    srwi 3, 3, 5
+; CHECK64-NEXT:    extsw 3, 3
+; CHECK64-NEXT:    blr
+entry:
+  tail call void @llvm.write_register.i32(metadata !0, i32 15)
+  %0 = load i32, ptr @mVal, align 4
+  %1 = tail call i32 @llvm.read_register.i32(metadata !0)
+  %cmp = icmp eq i32 %0, %1
+  %conv = zext i1 %cmp to i32
+  ret i32 %conv
+}
+
+declare i32 @llvm.read_register.i32(metadata)
+
+define dso_local void @testSetIntReg2(i32 noundef signext %xx) {
+; CHECK-LABEL: testSetIntReg2:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    stwu 1, -48(1)
+; CHECK-NEXT:    .cfi_def_cfa_offset 48
+; CHECK-NEXT:    .cfi_offset r23, -36
+; CHECK-NEXT:    stw 23, 12(1) # 4-byte Folded Spill
+; CHECK-NEXT:    mr 23, 3
+; CHECK-NEXT:    lwz 23, 12(1) # 4-byte Folded Reload
+; CHECK-NEXT:    addi 1, 1, 48
+; CHECK-NEXT:    blr
+;
+; CHECK64-LABEL: testSetIntReg2:
+; CHECK64:       # %bb.0: # %entry
+; CHECK64-NEXT:    std 23, -72(1) # 8-byte Folded Spill
+; CHECK64-NEXT:    mr 23, 3
+; CHECK64-NEXT:    ld 23, -72(1) # 8-byte Folded Reload
+; CHECK64-NEXT:    blr
+entry:
+  tail call void @llvm.write_register.i32(metadata !1, i32 %xx)
+  ret void
+}
+
+define dso_local signext i32 @testReturnReg() {
+; CHECK-LABEL: testReturnReg:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    stwu 1, -48(1)
+; CHECK-NEXT:    .cfi_def_cfa_offset 48
+; CHECK-NEXT:    .cfi_offset r23, -36
+; CHECK-NEXT:    stw 23, 12(1) # 4-byte Folded Spill
+; CHECK-NEXT:    li 23, 125
+; CHECK-NEXT:    mr 3, 23
+; CHECK-NEXT:    lwz 23, 12(1) # 4-byte Folded Reload
+; CHECK-NEXT:    addi 1, 1, 48
+; CHECK-NEXT:    blr
+;
+; CHECK64-LABEL: testReturnReg:
+; CHECK64:       # %bb.0: # %entry
+; CHECK64-NEXT:    std 23, -72(1) # 8-byte Folded Spill
+; CHECK64-NEXT:    li 23, 125
+; CHECK64-NEXT:    extsw 3, 23
+; CHECK64-NEXT:    ld 23, -72(1) # 8-byte Folded Reload
+; CHECK64-NEXT:    blr
+entry:
+  tail call void @llvm.write_register.i32(metadata !1, i32 125)
+  %0 = tail call i32 @llvm.read_register.i32(metadata !1)
+  ret i32 %0
+}
+
+define dso_local void @testViaASM(i32 noundef signext %xx) {
+; CHECK-LABEL: testViaASM:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    stwu 1, -64(1)
+; CHECK-NEXT:    .cfi_def_cfa_offset 64
+; CHECK-NEXT:    .cfi_offset r20, -48
+; CHECK-NEXT:    stw 20, 16(1) # 4-byte Folded Spill
+; CHECK-NEXT:    mr 20, 3
+; CHECK-NEXT:    #APP
+; CHECK-NEXT:    addi 3, 1, 1
+; CHECK-NEXT:    #NO_APP
+; CHECK-NEXT:    lis 4, myGVal@ha
+; CHECK-NEXT:    stw 3, myGVal@l(4)
+; CHECK-NEXT:    lwz 20, 16(1) # 4-byte Folded Reload
+; CHECK-NEXT:    addi 1, 1, 64
+; CHECK-NEXT:    blr
+;
+; CHECK64-LABEL: testViaASM:
+; CHECK64:       # %bb.0: # %entry
+; CHECK64-NEXT:    std 20, -96(1) # 8-byte Folded Spill
+; CHECK64-NEXT:    mr 20, 3
+; CHECK64-NEXT:    #APP
+; CHECK64-NEXT:    addi 3, 1, 1
+; CHECK64-NEXT:    #NO_APP
+; CHECK64-NEXT:    addis 4, 2, myGVal@toc@ha
+; CHECK64-NEXT:    addi 4, 4, myGVal@toc@l
+; CHECK64-NEXT:    stw 3, 0(4)
+; CHECK64-NEXT:    ld 20, -96(1) # 8-byte Folded Reload
+; CHECK64-NEXT:    blr
+entry:
+  tail call void @llvm.write_register.i32(metadata !2, i32 %xx)
+  %0 = tail call i32 @llvm.read_register.i32(metadata !2)
+  %1 = tail call i32 asm "addi $0, $2, $2", "=r,{r20},K"(i32 %0, i32 1)
+  store i32 %1, ptr @myGVal, align 4
+  ret void
+}
+
+!0 = !{!"r5"}
+!1 = !{!"r23"}
+!2 = !{!"r20"}