[llvm] dd36a69 - [NFC][Sparc] Pre-commit a test showing inefficient and broken LD/GD TLS

[Jessica Clarke via llvm-commits]

Author: Jessica Clarke
Date: 2025-07-22T16:43:30+01:00
New Revision: dd36a6901249796a175c3058a50d97cc3a7ba52c

URL: https://github.com/llvm/llvm-project/commit/dd36a6901249796a175c3058a50d97cc3a7ba52c
DIFF: https://github.com/llvm/llvm-project/commit/dd36a6901249796a175c3058a50d97cc3a7ba52c.diff

LOG: [NFC][Sparc] Pre-commit a test showing inefficient and broken LD/GD TLS

Added: 
    llvm/test/CodeGen/SPARC/tls-sp.ll

Modified: 
    

Removed: 
    


################################################################################
diff  --git a/llvm/test/CodeGen/SPARC/tls-sp.ll b/llvm/test/CodeGen/SPARC/tls-sp.ll
new file mode 100644
index 0000000000000..928d1dcbe8300
--- /dev/null
+++ b/llvm/test/CodeGen/SPARC/tls-sp.ll
@@ -0,0 +1,111 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -mtriple=sparc -relocation-model=pic < %s | FileCheck --check-prefix=SPARC %s
+; RUN: llc -mtriple=sparc64 -relocation-model=pic < %s | FileCheck --check-prefix=SPARC64 %s
+
+;; TODO: Fix the code generation for these functions.
+
+ at x = external thread_local global i8
+
+;; Test that we don't over-allocate stack space when calling __tls_get_addr
+;; with the call frame pseudos able to be eliminated.
+define ptr @no_alloca() nounwind {
+; SPARC-LABEL: no_alloca:
+; SPARC:       ! %bb.0: ! %entry
+; SPARC-NEXT:    save %sp, -96, %sp
+; SPARC-NEXT:  .Ltmp0:
+; SPARC-NEXT:    call .Ltmp1
+; SPARC-NEXT:  .Ltmp2:
+; SPARC-NEXT:    sethi %hi(_GLOBAL_OFFSET_TABLE_+(.Ltmp2-.Ltmp0)), %i0
+; SPARC-NEXT:  .Ltmp1:
+; SPARC-NEXT:    or %i0, %lo(_GLOBAL_OFFSET_TABLE_+(.Ltmp1-.Ltmp0)), %i0
+; SPARC-NEXT:    add %i0, %o7, %i0
+; SPARC-NEXT:    sethi %tgd_hi22(x), %i1
+; SPARC-NEXT:    add %i1, %tgd_lo10(x), %i1
+; SPARC-NEXT:    add %i0, %i1, %o0, %tgd_add(x)
+; SPARC-NEXT:    call __tls_get_addr, %tgd_call(x)
+; SPARC-NEXT:    nop
+; SPARC-NEXT:    ret
+; SPARC-NEXT:    restore %g0, %o0, %o0
+;
+; SPARC64-LABEL: no_alloca:
+; SPARC64:       ! %bb.0: ! %entry
+; SPARC64-NEXT:    save %sp, -144, %sp
+; SPARC64-NEXT:  .Ltmp0:
+; SPARC64-NEXT:    rd %pc, %o7
+; SPARC64-NEXT:  .Ltmp2:
+; SPARC64-NEXT:    sethi %hi(_GLOBAL_OFFSET_TABLE_+(.Ltmp2-.Ltmp0)), %i0
+; SPARC64-NEXT:  .Ltmp1:
+; SPARC64-NEXT:    or %i0, %lo(_GLOBAL_OFFSET_TABLE_+(.Ltmp1-.Ltmp0)), %i0
+; SPARC64-NEXT:    add %i0, %o7, %i0
+; SPARC64-NEXT:    sethi %tgd_hi22(x), %i1
+; SPARC64-NEXT:    add %i1, %tgd_lo10(x), %i1
+; SPARC64-NEXT:    add %i0, %i1, %o0, %tgd_add(x)
+; SPARC64-NEXT:    call __tls_get_addr, %tgd_call(x)
+; SPARC64-NEXT:    nop
+; SPARC64-NEXT:    ret
+; SPARC64-NEXT:    restore %g0, %o0, %o0
+entry:
+  %0 = call ptr @llvm.threadlocal.address.p0(ptr @x)
+  ret ptr %0
+}
+
+;; Test that %sp is valid for the call to __tls_get_addr. We store to a dynamic
+;; alloca in order to prevent eliminating any call frame pseudos from the call.
+define ptr @dynamic_alloca(i64 %n) nounwind {
+; SPARC-LABEL: dynamic_alloca:
+; SPARC:       ! %bb.0: ! %entry
+; SPARC-NEXT:    save %sp, -96, %sp
+; SPARC-NEXT:  .Ltmp3:
+; SPARC-NEXT:    call .Ltmp4
+; SPARC-NEXT:  .Ltmp5:
+; SPARC-NEXT:    sethi %hi(_GLOBAL_OFFSET_TABLE_+(.Ltmp5-.Ltmp3)), %i0
+; SPARC-NEXT:  .Ltmp4:
+; SPARC-NEXT:    or %i0, %lo(_GLOBAL_OFFSET_TABLE_+(.Ltmp4-.Ltmp3)), %i0
+; SPARC-NEXT:    add %i0, %o7, %i0
+; SPARC-NEXT:    add %sp, -1, %sp
+; SPARC-NEXT:    sethi %tgd_hi22(x), %i2
+; SPARC-NEXT:    add %i2, %tgd_lo10(x), %i2
+; SPARC-NEXT:    add %i0, %i2, %o0, %tgd_add(x)
+; SPARC-NEXT:    call __tls_get_addr, %tgd_call(x)
+; SPARC-NEXT:    nop
+; SPARC-NEXT:    add %sp, 1, %sp
+; SPARC-NEXT:    add %i1, 7, %i0
+; SPARC-NEXT:    and %i0, -8, %i0
+; SPARC-NEXT:    sub %sp, %i0, %i0
+; SPARC-NEXT:    add %i0, -8, %sp
+; SPARC-NEXT:    mov 1, %i1
+; SPARC-NEXT:    stb %i1, [%i0+88]
+; SPARC-NEXT:    ret
+; SPARC-NEXT:    restore %g0, %o0, %o0
+;
+; SPARC64-LABEL: dynamic_alloca:
+; SPARC64:       ! %bb.0: ! %entry
+; SPARC64-NEXT:    save %sp, -128, %sp
+; SPARC64-NEXT:  .Ltmp3:
+; SPARC64-NEXT:    rd %pc, %o7
+; SPARC64-NEXT:  .Ltmp5:
+; SPARC64-NEXT:    sethi %hi(_GLOBAL_OFFSET_TABLE_+(.Ltmp5-.Ltmp3)), %i1
+; SPARC64-NEXT:  .Ltmp4:
+; SPARC64-NEXT:    or %i1, %lo(_GLOBAL_OFFSET_TABLE_+(.Ltmp4-.Ltmp3)), %i1
+; SPARC64-NEXT:    add %i1, %o7, %i1
+; SPARC64-NEXT:    add %sp, -1, %sp
+; SPARC64-NEXT:    sethi %tgd_hi22(x), %i2
+; SPARC64-NEXT:    add %i2, %tgd_lo10(x), %i2
+; SPARC64-NEXT:    add %i1, %i2, %o0, %tgd_add(x)
+; SPARC64-NEXT:    call __tls_get_addr, %tgd_call(x)
+; SPARC64-NEXT:    nop
+; SPARC64-NEXT:    add %sp, 1, %sp
+; SPARC64-NEXT:    add %i0, 15, %i0
+; SPARC64-NEXT:    and %i0, -16, %i0
+; SPARC64-NEXT:    sub %sp, %i0, %i0
+; SPARC64-NEXT:    mov %i0, %sp
+; SPARC64-NEXT:    mov 1, %i1
+; SPARC64-NEXT:    stb %i1, [%i0+2175]
+; SPARC64-NEXT:    ret
+; SPARC64-NEXT:    restore %g0, %o0, %o0
+entry:
+  %0 = call ptr @llvm.threadlocal.address.p0(ptr @x)
+  %1 = alloca i8, i64 %n
+  store i8 1, ptr %1
+  ret ptr %0
+}