[llvm] 10d3c6b - [ShrinkWrap] Consider constant pool access as non-stack access (#164393)

[via llvm-commits]

Author: Sushant Gokhale
Date: 2025-10-22T04:05:57-07:00
New Revision: 10d3c6bc11ca919563036aa590440c38502bcd2f

URL: https://github.com/llvm/llvm-project/commit/10d3c6bc11ca919563036aa590440c38502bcd2f
DIFF: https://github.com/llvm/llvm-project/commit/10d3c6bc11ca919563036aa590440c38502bcd2f.diff

LOG: [ShrinkWrap] Consider constant pool access as non-stack access (#164393)

As far as I understand, constant pool access does not access stack and
accesses read-only memory.

This patch considers constant pool access as non-stack access allowing
shrink wrapping to happen in the concerned test.

We should be seeing perf improvement with povray benchmark from
SPEC17(around 12% with -flto -Ofast) after this patch.


An NFC PR #162476 already exists to upload the test before the patch but
approval has got delayed. So, as @davemgreen suggested in that PR, I
have uploaded the test and patch in this single PR to show how test
looks like.

Added: 
    

Modified: 
    llvm/lib/CodeGen/ShrinkWrap.cpp
    llvm/test/CodeGen/AArch64/shrink-wrap-const-pool-access.mir
    llvm/test/CodeGen/X86/fp128-select.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/CodeGen/ShrinkWrap.cpp b/llvm/lib/CodeGen/ShrinkWrap.cpp
index 826e4126de44c..83581052560cb 100644
--- a/llvm/lib/CodeGen/ShrinkWrap.cpp
+++ b/llvm/lib/CodeGen/ShrinkWrap.cpp
@@ -319,7 +319,7 @@ bool ShrinkWrapImpl::useOrDefCSROrFI(const MachineInstr &MI, RegScavenger *RS,
       return isa<GlobalValue>(UO);
     }
     if (const PseudoSourceValue *PSV = Op->getPseudoValue())
-      return PSV->isJumpTable();
+      return PSV->isJumpTable() || PSV->isConstantPool();
     return false;
   };
   // Load/store operations may access the stack indirectly when we previously

diff  --git a/llvm/test/CodeGen/AArch64/shrink-wrap-const-pool-access.mir b/llvm/test/CodeGen/AArch64/shrink-wrap-const-pool-access.mir
index 80312740607b2..6f33a75ab9fcb 100644
--- a/llvm/test/CodeGen/AArch64/shrink-wrap-const-pool-access.mir
+++ b/llvm/test/CodeGen/AArch64/shrink-wrap-const-pool-access.mir
@@ -21,12 +21,14 @@
     br label %common.ret
   }
 ...
-# FIXME: Following code has a load from constant pool. Accessing constant pool
+# Following code has a load from constant pool. Accessing constant pool
 # must not be considered as a stack access and hence, shrink wrapping must
 # happen.
-# CHECK-LABEL:name: shrink_wrap_load_from_const_pool
-# CHECK-NOT: savePoint
-# CHECK-NOT: restorePoint
+# CHECK-LABEL:name:            shrink_wrap_load_from_const_pool
+# CHECK:  savePoint:
+# CHECK:    - point:           '%bb.3'
+# CHECK:  restorePoint:
+# CHECK:    - point:           '%bb.5'
 ---
 name:            shrink_wrap_load_from_const_pool
 tracksRegLiveness: true

diff  --git a/llvm/test/CodeGen/X86/fp128-select.ll b/llvm/test/CodeGen/X86/fp128-select.ll
index 659e4ddedc646..27a651e23f886 100644
--- a/llvm/test/CodeGen/X86/fp128-select.ll
+++ b/llvm/test/CodeGen/X86/fp128-select.ll
@@ -13,8 +13,8 @@ define void @test_select(ptr %p, ptr %q, i1 zeroext %c) nounwind {
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    testl %edx, %edx
 ; SSE-NEXT:    jne .LBB0_1
-; SSE-NEXT:  # %bb.3:
-; SSE-NEXT:    movaps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; SSE-NEXT:  # %bb.2:
+; SSE-NEXT:    movaps {{.*#+}} xmm0 = [NaN]
 ; SSE-NEXT:    movaps %xmm0, (%rsi)
 ; SSE-NEXT:    retq
 ; SSE-NEXT:  .LBB0_1:
@@ -58,7 +58,7 @@ define fp128 @test_select_cc(fp128, fp128) nounwind {
 ; SSE-NEXT:    xorps %xmm1, %xmm1
 ; SSE-NEXT:    jmp .LBB1_3
 ; SSE-NEXT:  .LBB1_1:
-; SSE-NEXT:    movaps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
+; SSE-NEXT:    movaps {{.*#+}} xmm1 = [1.0E+0]
 ; SSE-NEXT:  .LBB1_3: # %BB0
 ; SSE-NEXT:    testl %ebx, %ebx
 ; SSE-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload