[llvm][AArch64] Add missing FPCR, H and B registers to Codeview mapping

[DavidSpickett]

Summary

Fixes https://github.com/llvm/llvm-project/issues/56484

H registers are 16 bit views of AArch64's Neon registers and
B are the 8 bit views.

msvc does not support 16 bit float (some mention in DirectX but I
couldn't find a way to get to it) so for lack of a better reference
I'm using:
https://github.com/MicrosoftEdge/JsDbg/blob/85c9b41b33bb8f3496dbe400d912c32bb7cc496b/server/references/dia/include/cvconst.h
(the other microsoft-pdb repo is no longer up to date)

Luckily clang does support fp16 so a test is added for that.

There is no 8 bit float type so I had to get creative with the
test case. We're not testing for correct debug info here just
that we can select the B register and not crash in the process.

For FPCR it's never going to be passed as an argument so I've
not added a test for it. It is included to keep our list looking
the same as the reference.

[DavidSpickett]

If you're not familiar with the naming: https://developer.arm.com/documentation/102374/0100/Registers-in-AArch64---general-purpose-registers

llvm does use the B registers, or at least can but usually for an instruction operand not for a variable you'd get debug info for. H was easy because __fp16 arguments end up there.

I think testing with MIR is the way to go to be after register allocation, and I did test that my hacks for the B test work with expensive checks. If there's a better way I'm happy to change it though.

[majnemer]

LGTM

llvm/include/llvm/DebugInfo/CodeView/CodeViewRegisters.def

 CV_REGISTER(ARM_NQ13, 413)
 CV_REGISTER(ARM_NQ14, 414)
 CV_REGISTER(ARM_NQ15, 415)
 
 #endif // defined(CV_REGISTERS_ALL) || defined(CV_REGISTERS_ARM)
 
 #if defined(CV_REGISTERS_ALL) || defined(CV_REGISTERS_ARM64)
 
-// arm64intr.h from MSVC defines ARM64_FPSR, which conflicts with
+// arm64intr.h from MSVC defines ARM64_FPSR and ARM64_FPCR, which conflicts with
 // these declarations.
 #pragma push_macro("ARM64_FPSR")
+#pragma push_macro("ARM64_FPCR")
 #undef ARM64_FPSR
+#undef ARM64_FPCR
 
 // ARM64 registers
 
 CV_REGISTER(ARM64_NOREG, 0)
 
 // General purpose 32-bit integer registers
 
 CV_REGISTER(ARM64_W0, 10)

 CV_REGISTER(ARM64_Q28, 208)
 CV_REGISTER(ARM64_Q29, 209)
 CV_REGISTER(ARM64_Q30, 210)
 CV_REGISTER(ARM64_Q31, 211)
 
 // Floating point status register
 
 CV_REGISTER(ARM64_FPSR, 220)
+CV_REGISTER(ARM64_FPCR, 221)
+
+// 8 bit floating point registers
+
+CV_REGISTER(ARM64_B0, 230)
+CV_REGISTER(ARM64_B1, 231)
+CV_REGISTER(ARM64_B2, 232)
+CV_REGISTER(ARM64_B3, 233)
+CV_REGISTER(ARM64_B4, 234)
+CV_REGISTER(ARM64_B5, 235)
+CV_REGISTER(ARM64_B6, 236)
+CV_REGISTER(ARM64_B7, 237)
+CV_REGISTER(ARM64_B8, 238)
+CV_REGISTER(ARM64_B9, 239)
+CV_REGISTER(ARM64_B10, 240)
+CV_REGISTER(ARM64_B11, 241)
+CV_REGISTER(ARM64_B12, 242)
+CV_REGISTER(ARM64_B13, 243)
+CV_REGISTER(ARM64_B14, 244)
+CV_REGISTER(ARM64_B15, 245)
+CV_REGISTER(ARM64_B16, 246)
+CV_REGISTER(ARM64_B17, 247)
+CV_REGISTER(ARM64_B18, 248)
+CV_REGISTER(ARM64_B19, 249)
+CV_REGISTER(ARM64_B20, 250)
+CV_REGISTER(ARM64_B21, 251)
+CV_REGISTER(ARM64_B22, 252)
+CV_REGISTER(ARM64_B23, 253)
+CV_REGISTER(ARM64_B24, 254)
+CV_REGISTER(ARM64_B25, 255)
+CV_REGISTER(ARM64_B26, 256)
+CV_REGISTER(ARM64_B27, 257)
+CV_REGISTER(ARM64_B28, 258)
+CV_REGISTER(ARM64_B29, 259)
+CV_REGISTER(ARM64_B30, 260)
+CV_REGISTER(ARM64_B31, 261)
+
+// 16 bit floating point registers
+
+CV_REGISTER(ARM64_H0, 270)
+CV_REGISTER(ARM64_H1, 271)
+CV_REGISTER(ARM64_H2, 272)
+CV_REGISTER(ARM64_H3, 273)
+CV_REGISTER(ARM64_H4, 274)
+CV_REGISTER(ARM64_H5, 275)
+CV_REGISTER(ARM64_H6, 276)
+CV_REGISTER(ARM64_H7, 277)
+CV_REGISTER(ARM64_H8, 278)
+CV_REGISTER(ARM64_H9, 279)
+CV_REGISTER(ARM64_H10, 280)
+CV_REGISTER(ARM64_H11, 281)
+CV_REGISTER(ARM64_H12, 282)
+CV_REGISTER(ARM64_H13, 283)
+CV_REGISTER(ARM64_H14, 284)
+CV_REGISTER(ARM64_H15, 285)
+CV_REGISTER(ARM64_H16, 286)
+CV_REGISTER(ARM64_H17, 287)
+CV_REGISTER(ARM64_H18, 288)
+CV_REGISTER(ARM64_H19, 289)
+CV_REGISTER(ARM64_H20, 290)
+CV_REGISTER(ARM64_H21, 291)
+CV_REGISTER(ARM64_H22, 292)
+CV_REGISTER(ARM64_H23, 293)
+CV_REGISTER(ARM64_H24, 294)
+CV_REGISTER(ARM64_H25, 295)
+CV_REGISTER(ARM64_H26, 296)
+CV_REGISTER(ARM64_H27, 297)
+CV_REGISTER(ARM64_H28, 298)
+CV_REGISTER(ARM64_H29, 299)
+CV_REGISTER(ARM64_H30, 300)
+CV_REGISTER(ARM64_H31, 301)
 
 #pragma pop_macro("ARM64_FPSR")
+#pragma pop_macro("ARM64_FPCR")
 
 #endif // defined(CV_REGISTERS_ALL) || defined(CV_REGISTERS_ARM64)

llvm/lib/Target/AArch64/MCTargetDesc/AArch64MCTargetDesc.cpp

MCPhysReg Reg;
       {codeview::RegisterId::ARM64_Q24, AArch64::Q24},
       {codeview::RegisterId::ARM64_Q25, AArch64::Q25},
       {codeview::RegisterId::ARM64_Q26, AArch64::Q26},
       {codeview::RegisterId::ARM64_Q27, AArch64::Q27},
       {codeview::RegisterId::ARM64_Q28, AArch64::Q28},
       {codeview::RegisterId::ARM64_Q29, AArch64::Q29},
       {codeview::RegisterId::ARM64_Q30, AArch64::Q30},
       {codeview::RegisterId::ARM64_Q31, AArch64::Q31},
-
+      {codeview::RegisterId::ARM64_B0, AArch64::B0},
+      {codeview::RegisterId::ARM64_B1, AArch64::B1},
+      {codeview::RegisterId::ARM64_B2, AArch64::B2},
+      {codeview::RegisterId::ARM64_B3, AArch64::B3},
+      {codeview::RegisterId::ARM64_B4, AArch64::B4},
+      {codeview::RegisterId::ARM64_B5, AArch64::B5},
+      {codeview::RegisterId::ARM64_B6, AArch64::B6},
+      {codeview::RegisterId::ARM64_B7, AArch64::B7},
+      {codeview::RegisterId::ARM64_B8, AArch64::B8},
+      {codeview::RegisterId::ARM64_B9, AArch64::B9},
+      {codeview::RegisterId::ARM64_B10, AArch64::B10},
+      {codeview::RegisterId::ARM64_B11, AArch64::B11},
+      {codeview::RegisterId::ARM64_B12, AArch64::B12},
+      {codeview::RegisterId::ARM64_B13, AArch64::B13},
+      {codeview::RegisterId::ARM64_B14, AArch64::B14},
+      {codeview::RegisterId::ARM64_B15, AArch64::B15},
+      {codeview::RegisterId::ARM64_B16, AArch64::B16},
+      {codeview::RegisterId::ARM64_B17, AArch64::B17},
+      {codeview::RegisterId::ARM64_B18, AArch64::B18},
+      {codeview::RegisterId::ARM64_B19, AArch64::B19},
+      {codeview::RegisterId::ARM64_B20, AArch64::B20},
+      {codeview::RegisterId::ARM64_B21, AArch64::B21},
+      {codeview::RegisterId::ARM64_B22, AArch64::B22},
+      {codeview::RegisterId::ARM64_B23, AArch64::B23},
+      {codeview::RegisterId::ARM64_B24, AArch64::B24},
+      {codeview::RegisterId::ARM64_B25, AArch64::B25},
+      {codeview::RegisterId::ARM64_B26, AArch64::B26},
+      {codeview::RegisterId::ARM64_B27, AArch64::B27},
+      {codeview::RegisterId::ARM64_B28, AArch64::B28},
+      {codeview::RegisterId::ARM64_B29, AArch64::B29},
+      {codeview::RegisterId::ARM64_B30, AArch64::B30},
+      {codeview::RegisterId::ARM64_B31, AArch64::B31},
+      {codeview::RegisterId::ARM64_H0, AArch64::H0},
+      {codeview::RegisterId::ARM64_H1, AArch64::H1},
+      {codeview::RegisterId::ARM64_H2, AArch64::H2},
+      {codeview::RegisterId::ARM64_H3, AArch64::H3},
+      {codeview::RegisterId::ARM64_H4, AArch64::H4},
+      {codeview::RegisterId::ARM64_H5, AArch64::H5},
+      {codeview::RegisterId::ARM64_H6, AArch64::H6},
+      {codeview::RegisterId::ARM64_H7, AArch64::H7},
+      {codeview::RegisterId::ARM64_H8, AArch64::H8},
+      {codeview::RegisterId::ARM64_H9, AArch64::H9},
+      {codeview::RegisterId::ARM64_H10, AArch64::H10},
+      {codeview::RegisterId::ARM64_H11, AArch64::H11},
+      {codeview::RegisterId::ARM64_H12, AArch64::H12},
+      {codeview::RegisterId::ARM64_H13, AArch64::H13},
+      {codeview::RegisterId::ARM64_H14, AArch64::H14},
+      {codeview::RegisterId::ARM64_H15, AArch64::H15},
+      {codeview::RegisterId::ARM64_H16, AArch64::H16},
+      {codeview::RegisterId::ARM64_H17, AArch64::H17},
+      {codeview::RegisterId::ARM64_H18, AArch64::H18},
+      {codeview::RegisterId::ARM64_H19, AArch64::H19},
+      {codeview::RegisterId::ARM64_H20, AArch64::H20},
+      {codeview::RegisterId::ARM64_H21, AArch64::H21},
+      {codeview::RegisterId::ARM64_H22, AArch64::H22},
+      {codeview::RegisterId::ARM64_H23, AArch64::H23},
+      {codeview::RegisterId::ARM64_H24, AArch64::H24},
+      {codeview::RegisterId::ARM64_H25, AArch64::H25},
+      {codeview::RegisterId::ARM64_H26, AArch64::H26},
+      {codeview::RegisterId::ARM64_H27, AArch64::H27},
+      {codeview::RegisterId::ARM64_H28, AArch64::H28},
+      {codeview::RegisterId::ARM64_H29, AArch64::H29},
+      {codeview::RegisterId::ARM64_H30, AArch64::H30},
+      {codeview::RegisterId::ARM64_H31, AArch64::H31},
   };
   for (const auto &I : RegMap)
     MRI->mapLLVMRegToCVReg(I.Reg, static_cast<int>(I.CVReg));
 }
 
 bool AArch64_MC::isQForm(const MCInst &MI, const MCInstrInfo *MCII) {
   const auto &FPR128 = AArch64MCRegisterClasses[AArch64::FPR128RegClassID];
   return llvm::any_of(MI, [&](const MCOperand &Op) {

llvm/test/DebugInfo/COFF/AArch64/codeview-b-register.mir

+# This test checks that we have a mapping between the B registers and their Codeview numbers.
+# There is no 8 bit float type so this is a modified version of codeview-h-register.mir to
+# have a B register in the location info but not the instructions. Which is incorrect
+# but in ways this test does not care about.
+#
+# RUN: llc %s -mtriple=arm64-windows -filetype=obj -o %t --start-after=unpack-mi-bundles
+# RUN: llvm-readobj --codeview %t | FileCheck %s --check-prefix=OBJ
+#
+# OBJ: LocalSym {
+# OBJ:   Kind: S_LOCAL (0x113E)
+# OBJ:   Type: __half (0x46)
+# OBJ:   Flags [ (0x1)
+# OBJ:     IsParameter (0x1)
+# OBJ:   ]
+# OBJ:   VarName: x
+# OBJ: }
+# OBJ: DefRangeRegisterSym {
+# OBJ:   Kind: S_DEFRANGE_REGISTER (0x1141)
+# OBJ:   Register: ARM64_B0 (0xE6)
+# OBJ:   MayHaveNoName: 0
+# OBJ:   LocalVariableAddrRange {
+# OBJ:     OffsetStart: .text+0x0
+# OBJ:     ISectStart: 0x0
+# OBJ:     Range: 0x4
+# OBJ:   }
+# OBJ: }
+#
+--- |
+  define internal fastcc i1 @test.fn(half %0) !dbg !4 {
+  Entry:
+    call void @llvm.dbg.value(metadata half %0, metadata !11, metadata !DIExpression()), !dbg !13
+    %1 = fcmp une half 0xH0000, %0, !dbg !14
+    ret i1 %1
+  }
+
+  ; Function Attrs: nocallback nofree nosync nounwind readnone speculatable willreturn
+  declare void @llvm.dbg.value(metadata, metadata, metadata) #0
+
+  attributes #0 = { nocallback nofree nosync nounwind readnone speculatable willreturn }
+
+  !llvm.module.flags = !{!0, !1}
+  !llvm.dbg.cu = !{!2}
+
+  !0 = !{i32 2, !"Debug Info Version", i32 3}
+  !1 = !{i32 2, !"CodeView", i32 1}
+  !2 = distinct !DICompileUnit(language: DW_LANG_C99, file: !3, isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug)
+  !3 = !DIFile(filename: "test", directory: ".")
+  !4 = distinct !DISubprogram(name: "fn", linkageName: "test.fn", scope: !5, file: !5, line: 32, type: !6, scopeLine: 32, flags: DIFlagStaticMember, spFlags: DISPFlagLocalToUnit | DISPFlagDefinition, unit: !2, retainedNodes: !10)
+  !5 = !DIFile(filename: "test.o", directory: ".")
+  !6 = !DISubroutineType(types: !7)
+  !7 = !{!8, !9}
+  !8 = !DIBasicType(name: "bool", size: 1, encoding: DW_ATE_boolean)
+  !9 = !DIBasicType(name: "f16", size: 16, encoding: DW_ATE_float)
+  !10 = !{!11}
+  !11 = !DILocalVariable(name: "x", arg: 1, scope: !12, file: !5, line: 32, type: !9)
+  !12 = distinct !DILexicalBlock(scope: !4, file: !5, line: 32, column: 1)
+  !13 = !DILocation(line: 32, column: 31, scope: !12)
+  !14 = !DILocation(line: 33, column: 5, scope: !15)
+  !15 = distinct !DILexicalBlock(scope: !12, file: !5)
+...
+---
+name: test.fn
+body: |
+  bb.0:
+  liveins: $b0
+  DBG_VALUE $b0, $noreg, !11, !DIExpression(), debug-location !13
+  renamable $s0 = nofpexcept FCVTSHr killed renamable $h0, debug-location !14
+  DBG_VALUE $b0, $noreg, !11, !DIExpression(DW_OP_LLVM_entry_value, 1), debug-location !13
+  nofpexcept FCMPSri killed renamable $s0, implicit-def $nzcv, debug-location !14
+  renamable $w0 = CSINCWr $wzr, $wzr, 0, implicit killed $nzcv, debug-location !14
+  RET undef $lr, implicit killed $w0
+...

llvm/test/DebugInfo/COFF/AArch64/codeview-h-register.mir

+# This test checks that we have a mapping between the H register chosen by codegen
+# and the codeview number for that register.
+#
+# RUN: llc %s -mtriple=arm64-windows -filetype=obj -o %t --start-after=unpack-mi-bundles
+# RUN: llvm-readobj --codeview %t | FileCheck %s --check-prefix=OBJ
+#
+# OBJ: LocalSym {
+# OBJ:   Kind: S_LOCAL (0x113E)
+# OBJ:   Type: __half (0x46)
+# OBJ:   Flags [ (0x1)
+# OBJ:     IsParameter (0x1)
+# OBJ:   ]
+# OBJ:   VarName: x
+# OBJ: }
+# OBJ: DefRangeRegisterSym {
+# OBJ:   Kind: S_DEFRANGE_REGISTER (0x1141)
+# OBJ:   Register: ARM64_H0 (0x10E)
+# OBJ:   MayHaveNoName: 0
+# OBJ:   LocalVariableAddrRange {
+# OBJ:     OffsetStart: .text+0x0
+# OBJ:     ISectStart: 0x0
+# OBJ:     Range: 0x4
+# OBJ:   }
+# OBJ: }
+#
+--- |
+  define internal fastcc i1 @test.fn(half %0) !dbg !4 {
+  Entry:
+    call void @llvm.dbg.value(metadata half %0, metadata !11, metadata !DIExpression()), !dbg !13
+    %1 = fcmp une half 0xH0000, %0, !dbg !14
+    ret i1 %1
+  }
+
+  ; Function Attrs: nocallback nofree nosync nounwind readnone speculatable willreturn
+  declare void @llvm.dbg.value(metadata, metadata, metadata) #0
+
+  attributes #0 = { nocallback nofree nosync nounwind readnone speculatable willreturn }
+
+  !llvm.module.flags = !{!0, !1}
+  !llvm.dbg.cu = !{!2}
+
+  !0 = !{i32 2, !"Debug Info Version", i32 3}
+  !1 = !{i32 2, !"CodeView", i32 1}
+  !2 = distinct !DICompileUnit(language: DW_LANG_C99, file: !3, isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug)
+  !3 = !DIFile(filename: "test", directory: ".")
+  !4 = distinct !DISubprogram(name: "fn", linkageName: "test.fn", scope: !5, file: !5, line: 32, type: !6, scopeLine: 32, flags: DIFlagStaticMember, spFlags: DISPFlagLocalToUnit | DISPFlagDefinition, unit: !2, retainedNodes: !10)
+  !5 = !DIFile(filename: "test.o", directory: ".")
+  !6 = !DISubroutineType(types: !7)
+  !7 = !{!8, !9}
+  !8 = !DIBasicType(name: "bool", size: 1, encoding: DW_ATE_boolean)
+  !9 = !DIBasicType(name: "f16", size: 16, encoding: DW_ATE_float)
+  !10 = !{!11}
+  !11 = !DILocalVariable(name: "x", arg: 1, scope: !12, file: !5, line: 32, type: !9)
+  !12 = distinct !DILexicalBlock(scope: !4, file: !5, line: 32, column: 1)
+  !13 = !DILocation(line: 32, column: 31, scope: !12)
+  !14 = !DILocation(line: 33, column: 5, scope: !15)
+  !15 = distinct !DILexicalBlock(scope: !12, file: !5)
+...
+---
+name: test.fn
+body: |
+  bb.0:
+  liveins: $h0
+  DBG_VALUE $h0, $noreg, !11, !DIExpression(), debug-location !13
+  renamable $s0 = nofpexcept FCVTSHr killed renamable $h0, debug-location !14
+  DBG_VALUE $h0, $noreg, !11, !DIExpression(DW_OP_LLVM_entry_value, 1), debug-location !13
+  nofpexcept FCMPSri killed renamable $s0, implicit-def $nzcv, debug-location !14
+  renamable $w0 = CSINCWr $wzr, $wzr, 0, implicit killed $nzcv, debug-location !14
+  RET undef $lr, implicit killed $w0
+...