[llvm] 4b99eb2 - [Sparc] Remove bogus stack adjustment for LD/GD TLS (#149890)

via llvm-commits llvm-commits at lists.llvm.org
Tue Jul 22 08:44:21 PDT 2025


Author: Jessica Clarke
Date: 2025-07-22T16:44:18+01:00
New Revision: 4b99eb2eb49d8fcdb29ba494501e481cf09831ae

URL: https://github.com/llvm/llvm-project/commit/4b99eb2eb49d8fcdb29ba494501e481cf09831ae
DIFF: https://github.com/llvm/llvm-project/commit/4b99eb2eb49d8fcdb29ba494501e481cf09831ae.diff

LOG: [Sparc] Remove bogus stack adjustment for LD/GD TLS (#149890)

This argument is the number of bytes to adjust the stack by for the
duration of the call. In most cases, PEI is able to eliminate the
corresponding call frame pseudos, folding them into the initial stack
frame allocation (rounded up to stack alignment), where it just ends up
allocating more space than needed. However, in the rare case where this
cannot be done, e.g. due to the use of a dynamic alloca, the 1 byte
stack adjustment persists and results in a misaligned stack for the
duration of the call. This has been the case ever since TLS support was
added in cb1dca602c43 ("[Sparc] Add support for TLS in sparc."), and I
can only assume that 1 was used erroneously thinking that it is the
number of arguments (as there is 1 register argument for the call), not
the number of bytes for on-stack arguments.

Fixes: https://github.com/llvm/llvm-project/issues/149808

Added: 
    

Modified: 
    llvm/lib/Target/Sparc/SparcISelLowering.cpp
    llvm/test/CodeGen/SPARC/tls-sp.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/Sparc/SparcISelLowering.cpp b/llvm/lib/Target/Sparc/SparcISelLowering.cpp
index 9b434d87c2676..1aa8efe3e9979 100644
--- a/llvm/lib/Target/Sparc/SparcISelLowering.cpp
+++ b/llvm/lib/Target/Sparc/SparcISelLowering.cpp
@@ -2201,7 +2201,7 @@ SDValue SparcTargetLowering::LowerGlobalTLSAddress(SDValue Op,
     SDValue Chain = DAG.getEntryNode();
     SDValue InGlue;
 
-    Chain = DAG.getCALLSEQ_START(Chain, 1, 0, DL);
+    Chain = DAG.getCALLSEQ_START(Chain, 0, 0, DL);
     Chain = DAG.getCopyToReg(Chain, DL, SP::O0, Argument, InGlue);
     InGlue = Chain.getValue(1);
     SDValue Callee = DAG.getTargetExternalSymbol("__tls_get_addr", PtrVT);
@@ -2219,7 +2219,7 @@ SDValue SparcTargetLowering::LowerGlobalTLSAddress(SDValue Op,
                      InGlue};
     Chain = DAG.getNode(SPISD::TLS_CALL, DL, NodeTys, Ops);
     InGlue = Chain.getValue(1);
-    Chain = DAG.getCALLSEQ_END(Chain, 1, 0, InGlue, DL);
+    Chain = DAG.getCALLSEQ_END(Chain, 0, 0, InGlue, DL);
     InGlue = Chain.getValue(1);
     SDValue Ret = DAG.getCopyFromReg(Chain, DL, SP::O0, PtrVT, InGlue);
 

diff  --git a/llvm/test/CodeGen/SPARC/tls-sp.ll b/llvm/test/CodeGen/SPARC/tls-sp.ll
index 928d1dcbe8300..de9af01398d23 100644
--- a/llvm/test/CodeGen/SPARC/tls-sp.ll
+++ b/llvm/test/CodeGen/SPARC/tls-sp.ll
@@ -2,8 +2,6 @@
 ; 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.
-
 @x = external thread_local global i8
 
 ;; Test that we don't over-allocate stack space when calling __tls_get_addr
@@ -29,7 +27,7 @@ define ptr @no_alloca() nounwind {
 ;
 ; SPARC64-LABEL: no_alloca:
 ; SPARC64:       ! %bb.0: ! %entry
-; SPARC64-NEXT:    save %sp, -144, %sp
+; SPARC64-NEXT:    save %sp, -128, %sp
 ; SPARC64-NEXT:  .Ltmp0:
 ; SPARC64-NEXT:    rd %pc, %o7
 ; SPARC64-NEXT:  .Ltmp2:
@@ -62,13 +60,11 @@ define ptr @dynamic_alloca(i64 %n) nounwind {
 ; 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
@@ -88,13 +84,11 @@ define ptr @dynamic_alloca(i64 %n) nounwind {
 ; 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


        


More information about the llvm-commits mailing list