[llvm] r363418 - [AMDGPU] Don't constrain callees with inlinehint from inlining on MaxBB check

Valery Pykhtin via llvm-commits llvm-commits at lists.llvm.org
Fri Jun 14 09:37:33 PDT 2019


Author: vpykhtin
Date: Fri Jun 14 09:37:33 2019
New Revision: 363418

URL: http://llvm.org/viewvc/llvm-project?rev=363418&view=rev
Log:
[AMDGPU] Don't constrain callees with inlinehint from inlining on MaxBB check

Summary: Function bodies marked inline in an opencl source are eliminated but MaxBB check may prevent inlining them leaving undefined references.

Reviewers: rampitec, arsenm

Subscribers: kzhuravl, jvesely, wdng, nhaehnle, yaxunl, dstuttard, tpr, Anastasia, t-tye, hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D63337

Modified:
    llvm/trunk/lib/Target/AMDGPU/AMDGPUInline.cpp
    llvm/trunk/test/CodeGen/AMDGPU/inline-maxbb.ll

Modified: llvm/trunk/lib/Target/AMDGPU/AMDGPUInline.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/AMDGPU/AMDGPUInline.cpp?rev=363418&r1=363417&r2=363418&view=diff
==============================================================================
--- llvm/trunk/lib/Target/AMDGPU/AMDGPUInline.cpp (original)
+++ llvm/trunk/lib/Target/AMDGPU/AMDGPUInline.cpp Fri Jun 14 09:37:33 2019
@@ -218,7 +218,7 @@ InlineCost AMDGPUInliner::getInlineCost(
                              LocalParams, TTI, GetAssumptionCache, None, PSI,
                              RemarksEnabled ? &ORE : nullptr);
 
-  if (IC && !IC.isAlways()) {
+  if (IC && !IC.isAlways() && !Callee->hasFnAttribute(Attribute::InlineHint)) {
     // Single BB does not increase total BB amount, thus subtract 1
     size_t Size = Caller->size() + Callee->size() - 1;
     if (MaxBB && Size > MaxBB)

Modified: llvm/trunk/test/CodeGen/AMDGPU/inline-maxbb.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/inline-maxbb.ll?rev=363418&r1=363417&r2=363418&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/inline-maxbb.ll (original)
+++ llvm/trunk/test/CodeGen/AMDGPU/inline-maxbb.ll Fri Jun 14 09:37:33 2019
@@ -31,3 +31,38 @@ define amdgpu_kernel void @caller(i32 %x
   store volatile i32 %res, i32 addrspace(1)* undef
   ret void
 }
+
+
+; inlinehint
+define i32 @callee_hint(i32 %x) #0 {
+entry:
+  %cc = icmp eq i32 %x, 1
+  br i1 %cc, label %ret_res, label %mulx
+
+mulx:
+  %mul1 = mul i32 %x, %x
+  %mul2 = mul i32 %mul1, %x
+  %mul3 = mul i32 %mul1, %mul2
+  %mul4 = mul i32 %mul3, %mul2
+  %mul5 = mul i32 %mul4, %mul3
+  br label %ret_res
+
+ret_res:
+  %r = phi i32 [ %mul5, %mulx ], [ %x, %entry ]
+  ret i32 %r
+}
+
+; INL-LABEL: @caller_hint
+; NOINL-LABEL: @caller_hint
+; INL: mul i32
+; INL-NOT: call i32
+; NOINL: mul i32
+; NOINL-NOT: call i32
+
+define amdgpu_kernel void @caller_hint(i32 %x) {
+  %res = call i32 @callee_hint(i32 %x)
+  store volatile i32 %res, i32 addrspace(1)* undef
+  ret void
+}
+
+attributes #0 = { inlinehint }




More information about the llvm-commits mailing list