[AVR] Disallow the LDDWRdPtrQ instruction with Z as the destination

This is an AVR-specific workaround for a limitation of the register
allocator that only exposes itself on targets with high register
contention like AVR, which only has three pointer registers.

The three pointer registers are X, Y, and Z.
In most nontrivial functions, Y is reserved for the frame pointer,
as per the calling convention. This leaves X and Z. Some instructions,
such as LPM ("load program memory"), are only defined for the Z
register. Sometimes this just leaves X.

When the backend generates a LDDWRdPtrQ instruction with Z as the
destination pointer, it usually trips up the register allocator
with this error message:

  LLVM ERROR: ran out of registers during register allocation

This patch is a hacky workaround. We ban the LDDWRdPtrQ instruction
from ever using the Z register as an operand. This gives the
register allocator a bit more space to allocate, fixing the
regalloc exhaustion error.

Here is a description from the patch author Peter Nimmervoll

  As far as I understand the problem occurs when LDDWRdPtrQ uses
  the ptrdispregs register class as target register. This should work, but
  the allocator can't deal with this for some reason. So from my testing,
  it seams like (and I might be totally wrong on this) the allocator reserves
  the Z register for the ICALL instruction and then the register class
  ptrdispregs only has 1 register left and we can't use Y for source and
  destination. Removing the Z register from DREGS fixes the problem but
  removing Y register does not.

More information about the bug can be found on the avr-rust issue
tracker at avr-rust/rust-legacy-fork#37.

A bug has raised to track the removal of this workaround and a proper
fix; PR39553 at https://bugs.llvm.org/show_bug.cgi?id=39553.

Patch by Peter Nimmervoll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@346114 91177308-0d34-0410-b5e6-96231b3b80d8

lib/Target/AVR/AVRInstrInfo.td

@@ -1222,7 +1222,7 @@ isReMaterializable = 1 in
   // ldd Rd,   P+q
   // ldd Rd+1, P+q+1
   let Constraints = "@earlyclobber $dst" in
-  def LDDWRdPtrQ : Pseudo<(outs DREGS:$dst),
+  def LDDWRdPtrQ : Pseudo<(outs DREGS_WITHOUT_Z_WORKAROUND:$dst),
                           (ins memri:$memri),
                           "lddw\t$dst, $memri",
                           [(set i16:$dst, (load addr:$memri))]>,

lib/Target/AVR/AVRRegisterInfo.td

@@ -157,6 +157,26 @@ def DREGS : RegisterClass<"AVR", [i16], 8,
     R9R8, R7R6, R5R4, R3R2, R1R0
   )>;
 
+// The 16-bit DREGS register class, excluding the Z pointer register.
+//
+// This is used by instructions which cause high pointer register
+// contention which leads to an assertion in the register allocator.
+//
+// There is no technical reason why instructions that use this class
+// cannot use Z; it's simply a workaround a regalloc bug.
+//
+// More information can be found in PR39553.
+def DREGS_WITHOUT_Z_WORKAROUND : RegisterClass<"AVR", [i16], 8,
+  (
+    // Return value and arguments.
+    add R25R24, R19R18, R21R20, R23R22,
+    // Scratch registers.
+    R27R26,
+    // Callee saved registers.
+    R29R28, R17R16, R15R14, R13R12, R11R10,
+    R9R8, R7R6, R5R4, R3R2, R1R0
+  )>;
+
 // 16-bit register class for immediate instructions.
 def DLDREGS : RegisterClass<"AVR", [i16], 8,
   (

test/CodeGen/AVR/pseudo/LDDWRdPtrQ-same-src-dst.mir

@@ -25,11 +25,11 @@ body: |
 
     ; CHECK-LABEL: test_lddwrdptrq
 
-    ; CHECK:      ldd [[SCRATCH:r[0-9]+]], Z+10
+    ; CHECK:      ldd [[SCRATCH:r[0-9]+]], Y+10
     ; CHECK-NEXT: push [[SCRATCH]]
-    ; CHECK-NEXT: ldd [[SCRATCH]], Z+11
-    ; CHECK-NEXT: mov r31, [[SCRATCH]]
-    ; CHECK-NEXT: pop r30
+    ; CHECK-NEXT: ldd [[SCRATCH]], Y+11
+    ; CHECK-NEXT: mov r29, [[SCRATCH]]
+    ; CHECK-NEXT: pop r28
 
-    early-clobber $r31r30 = LDDWRdPtrQ undef $r31r30, 10
+    early-clobber $r29r28 = LDDWRdPtrQ undef $r29r28, 10
 ...

test/CodeGen/AVR/pseudo/LDDWRdPtrQ.mir

@@ -18,8 +18,8 @@ body: |
 
     ; CHECK-LABEL: test_lddwrdptrq
 
-    ; CHECK:      ldd     r30, Y+10
-    ; CHECK-NEXT: ldd     r31, Y+11
+    ; CHECK:      ldd     r28, Z+10
+    ; CHECK-NEXT: ldd     r29, Z+11
 
-    early-clobber $r31r30 = LDDWRdPtrQ undef $r29r28, 10
+    early-clobber $r29r28 = LDDWRdPtrQ undef $r31r30, 10
 ...

test/CodeGen/AVR/rust-avr-bug-37.ll

@@ -0,0 +1,25 @@
+; RUN: llc < %s -march=avr | FileCheck %s
+
+%"fmt::Formatter" = type { i32, { i8*, void (i8*)** } }
+
+@str.1b = external constant [0 x i8]
+
+define void @"TryFromIntError::Debug"(%"fmt::Formatter"* dereferenceable(32)) unnamed_addr #0 personality i32 (...)* @rust_eh_personality {
+; CHECK-LABEL: "TryFromIntError::Debug"
+start:
+  %builder = alloca i8, align 8
+  %1 = getelementptr inbounds %"fmt::Formatter", %"fmt::Formatter"* %0, i16 0, i32 1
+  %2 = bitcast { i8*, void (i8*)** }* %1 to {}**
+  %3 = load {}*, {}** %2, align 2
+  %4 = getelementptr inbounds %"fmt::Formatter", %"fmt::Formatter"* %0, i16 0, i32 1, i32 1
+  %5 = load void (i8*)**, void (i8*)*** %4, align 2
+  %6 = getelementptr inbounds void (i8*)*, void (i8*)** %5, i16 3
+  %7 = bitcast void (i8*)** %6 to i8 ({}*, i8*, i16)**
+  %8 = load i8 ({}*, i8*, i16)*, i8 ({}*, i8*, i16)** %7, align 2
+  %9 = tail call i8 %8({}* nonnull %3, i8* noalias nonnull readonly getelementptr inbounds ([0 x i8], [0 x i8]* @str.1b, i16 0, i16 0), i16 15)
+  unreachable
+}
+
+declare i32 @rust_eh_personality(...) unnamed_addr
+
+attributes #0 = { uwtable }

test/CodeGen/AVR/rust-avr-bug-95.ll

@@ -0,0 +1,37 @@
+; RUN: llc < %s -march=avr | FileCheck %s
+
+%"fmt::Formatter.1.77.153.229.305.381.1673" = type { [0 x i8], i32, [0 x i8], i32, [0 x i8], i8, [0 x i8], %"option::Option<usize>.0.76.152.228.304.380.1672", [0 x i8], %"option::Option<usize>.0.76.152.228.304.380.1672", [0 x i8], { {}*, {}* }, [0 x i8], { i8*, i8* }, [0 x i8], { [0 x { i8*, i8* }]*, i16 }, [0 x i8] }
+%"option::Option<usize>.0.76.152.228.304.380.1672" = type { [0 x i8], i8, [2 x i8] }
+
+@str.4S = external constant [5 x i8]
+
+; Function Attrs: uwtable
+define void @"_ZN65_$LT$lib..str..Chars$LT$$u27$a$GT$$u20$as$u20$lib..fmt..Debug$GT$3fmt17h76a537e22649f739E"(%"fmt::Formatter.1.77.153.229.305.381.1673"* dereferenceable(27) %__arg_0) unnamed_addr #0 personality i32 (...)* @rust_eh_personality {
+; CHECK-LABEL: "_ZN65_$LT$lib..str..Chars$LT$$u27$a$GT$$u20$as$u20$lib..fmt..Debug$GT$3fmt17h76a537e22649f739E"
+start:
+  %0 = getelementptr inbounds %"fmt::Formatter.1.77.153.229.305.381.1673", %"fmt::Formatter.1.77.153.229.305.381.1673"* %__arg_0, i16 0, i32 11, i32 0
+  %1 = load {}*, {}** %0, align 1, !noalias !0, !nonnull !9
+  %2 = getelementptr inbounds %"fmt::Formatter.1.77.153.229.305.381.1673", %"fmt::Formatter.1.77.153.229.305.381.1673"* %__arg_0, i16 0, i32 11, i32 1
+  %3 = bitcast {}** %2 to i1 ({}*, [0 x i8]*, i16)***
+  %4 = load i1 ({}*, [0 x i8]*, i16)**, i1 ({}*, [0 x i8]*, i16)*** %3, align 1, !noalias !0, !nonnull !9
+  %5 = getelementptr inbounds i1 ({}*, [0 x i8]*, i16)*, i1 ({}*, [0 x i8]*, i16)** %4, i16 3
+  %6 = load i1 ({}*, [0 x i8]*, i16)*, i1 ({}*, [0 x i8]*, i16)** %5, align 1, !invariant.load !9, !noalias !0, !nonnull !9
+  %7 = tail call zeroext i1 %6({}* nonnull %1, [0 x i8]* noalias nonnull readonly bitcast ([5 x i8]* @str.4S to [0 x i8]*), i16 5), !noalias !10
+  unreachable
+}
+
+declare i32 @rust_eh_personality(...) unnamed_addr
+
+attributes #0 = { uwtable }
+
+!0 = !{!1, !3, !5, !6, !8}
+!1 = distinct !{!1, !2, !"_ZN3lib3fmt9Formatter9write_str17ha1a9656fc66ccbe5E: %data.0"}
+!2 = distinct !{!2, !"_ZN3lib3fmt9Formatter9write_str17ha1a9656fc66ccbe5E"}
+!3 = distinct !{!3, !4, !"_ZN3lib3fmt8builders16debug_struct_new17h352a1de8f89c2bc3E: argument 0"}
+!4 = distinct !{!4, !"_ZN3lib3fmt8builders16debug_struct_new17h352a1de8f89c2bc3E"}
+!5 = distinct !{!5, !4, !"_ZN3lib3fmt8builders16debug_struct_new17h352a1de8f89c2bc3E: %name.0"}
+!6 = distinct !{!6, !7, !"_ZN3lib3fmt9Formatter12debug_struct17ha1ff79f633171b68E: argument 0"}
+!7 = distinct !{!7, !"_ZN3lib3fmt9Formatter12debug_struct17ha1ff79f633171b68E"}
+!8 = distinct !{!8, !7, !"_ZN3lib3fmt9Formatter12debug_struct17ha1ff79f633171b68E: %name.0"}
+!9 = !{}
+!10 = !{!3, !6}