[llvm] [PowerPC] Spill non-volatile registers required for traceback table (PR #71115)

[Chen Zheng via llvm-commits]

================
@@ -20,33 +21,66 @@ liveins:
 body:             |
   bb.0.entry:
     liveins: $x3
+    ; SAVEONE-LABEL: name: CRAllSave
+    ; SAVEONE: liveins: $x3, $x29, $cr2, $cr4
+    ; SAVEONE-NEXT: {{  $}}
+    ; SAVEONE-NEXT: $x12 = MFCR8 implicit killed $cr2, implicit killed $cr4
+    ; SAVEONE-NEXT: STW8 killed $x12, 8, $x1
+    ; SAVEONE-NEXT: STD killed $x29, -24, $x1 :: (store (s64) into %fixed-stack.0)
+    ; SAVEONE-NEXT: renamable $x29 = ANDI8_rec killed renamable $x3, 1, implicit-def dead $cr0, implicit-def $cr0gt
+    ; SAVEONE-NEXT: renamable $cr2lt = COPY $cr0gt
+    ; SAVEONE-NEXT: renamable $cr4lt = COPY $cr0gt
+    ; SAVEONE-NEXT: renamable $x3 = COPY $x29
+    ; SAVEONE-NEXT: $x29 = LD -24, $x1 :: (load (s64) from %fixed-stack.0)
+    ; SAVEONE-NEXT: $x12 = LWZ8 8, $x1
+    ; SAVEONE-NEXT: $cr2 = MTOCRF8 $x12
+    ; SAVEONE-NEXT: $cr4 = MTOCRF8 killed $x12
+    ; SAVEONE-NEXT: BLR8 implicit $lr8, implicit $rm, implicit $x3
+    ;
+    ; SAVEALL-LABEL: name: CRAllSave
+    ; SAVEALL: liveins: $x3, $x29, $cr2, $cr4
+    ; SAVEALL-NEXT: {{  $}}
+    ; SAVEALL-NEXT: $x12 = MFCR8 implicit killed $cr2, implicit killed $cr4
+    ; SAVEALL-NEXT: STW8 killed $x12, 8, $x1
+    ; SAVEALL-NEXT: STD killed $x29, -24, $x1 :: (store (s64) into %fixed-stack.0)
+    ; SAVEALL-NEXT: renamable $x29 = ANDI8_rec killed renamable $x3, 1, implicit-def dead $cr0, implicit-def $cr0gt
+    ; SAVEALL-NEXT: renamable $cr2lt = COPY $cr0gt
+    ; SAVEALL-NEXT: renamable $cr4lt = COPY $cr0gt
+    ; SAVEALL-NEXT: renamable $x3 = COPY $x29
+    ; SAVEALL-NEXT: $x29 = LD -24, $x1 :: (load (s64) from %fixed-stack.0)
+    ; SAVEALL-NEXT: $x12 = LWZ8 8, $x1
+    ; SAVEALL-NEXT: $cr2 = MTOCRF8 $x12
+    ; SAVEALL-NEXT: $cr4 = MTOCRF8 killed $x12
+    ; SAVEALL-NEXT: BLR8 implicit $lr8, implicit $rm, implicit $x3
+    ;
+    ; SAVEALL-AIX-LABEL: name: CRAllSave
+    ; SAVEALL-AIX: liveins: $x3, $x29, $x30, $x31, $cr2, $cr4
+    ; SAVEALL-AIX-NEXT: {{  $}}
+    ; SAVEALL-AIX-NEXT: $x12 = MFCR8 implicit killed $cr2, implicit killed $cr4
+    ; SAVEALL-AIX-NEXT: STW8 killed $x12, 8, $x1
+    ; SAVEALL-AIX-NEXT: STD killed $x29, -24, $x1 :: (store (s64) into %fixed-stack.2)
+    ; SAVEALL-AIX-NEXT: STD killed $x30, -16, $x1 :: (store (s64) into %fixed-stack.1, align 16)
+    ; SAVEALL-AIX-NEXT: STD killed $x31, -8, $x1 :: (store (s64) into %fixed-stack.0)
+    ; SAVEALL-AIX-NEXT: renamable $x29 = ANDI8_rec killed renamable $x3, 1, implicit-def dead $cr0, implicit-def $cr0gt
+    ; SAVEALL-AIX-NEXT: renamable $cr2lt = COPY $cr0gt
+    ; SAVEALL-AIX-NEXT: renamable $cr4lt = COPY $cr0gt
+    ; SAVEALL-AIX-NEXT: renamable $x3 = COPY $x29
+    ; SAVEALL-AIX-NEXT: $x31 = LD -8, $x1 :: (load (s64) from %fixed-stack.0)
+    ; SAVEALL-AIX-NEXT: $x30 = LD -16, $x1 :: (load (s64) from %fixed-stack.1, align 16)
+    ; SAVEALL-AIX-NEXT: $x29 = LD -24, $x1 :: (load (s64) from %fixed-stack.2)
+    ; SAVEALL-AIX-NEXT: $x12 = LWZ8 8, $x1
+    ; SAVEALL-AIX-NEXT: $cr2 = MTOCRF8 $x12
+    ; SAVEALL-AIX-NEXT: $cr4 = MTOCRF8 killed $x12
+    ; SAVEALL-AIX-NEXT: BLR8 implicit $lr8, implicit $rm, implicit $x3
----------------
chenzheng1030 wrote:

What I meant is this case is to test non-volatile CRs. So maybe we don't need to test the changes for non-volatile GPRs. Now because of the non-volatile GPR r29 input, the case is changed a lot. But the change is not for the original test point.

So to make the case focus on the code generation for non-volatile CRs, we can change the `r29` to `r3` to avoid the diffs for non-volatile GPRs. And we already have many cases for the non-volatile GPRs before.

What do you think?

https://github.com/llvm/llvm-project/pull/71115