AMDGPU: Uniform branch conditions can originate with intrinsics

Summary:
Discovered by Dave Airlie, fixes an assertion in Khronos OpenGL CTS
GL43-CTS.shader_storage_buffer_object.advanced-matrix.

In this particular case, the buffer load intrinsic fed into a uniform
conditional branch, and led the brcond lowering down the wrong path.

Reviewers: tstellarAMD, arsenm

Subscribers: arsenm, llvm-commits

Differential Revision: http://reviews.llvm.org/D19931

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268650 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/AMDGPU/SIISelLowering.cpp b/lib/Target/AMDGPU/SIISelLowering.cpp
index 21ab9de79c5b50238b926658a345c5a7f82d38ac..3ca39a7c8e581938ce99b6f1400ef5737ade8e2c 100644 (file)
--- a/lib/Target/AMDGPU/SIISelLowering.cpp
+++ b/lib/Target/AMDGPU/SIISelLowering.cpp
@@ -1356,14 +1356,13 @@ SDValue SITargetLowering::LowerBRCOND(SDValue BRCOND,
     Target = BR->getOperand(1);
   }
 
-  if (Intr->getOpcode() != ISD::INTRINSIC_W_CHAIN) {
+  if (!isCFIntrinsic(Intr)) {
     // This is a uniform branch so we don't need to legalize.
     return BRCOND;
   }
 
   assert(!SetCC ||
         (SetCC->getConstantOperandVal(1) == 1 &&
-         isCFIntrinsic(Intr) &&
          cast<CondCodeSDNode>(SetCC->getOperand(2).getNode())->get() ==
                                                              ISD::SETNE));
 

diff --git a/test/CodeGen/AMDGPU/uniform-branch-intrinsic-cond.ll b/test/CodeGen/AMDGPU/uniform-branch-intrinsic-cond.ll
new file mode 100644 (file)
index 0000000..93a2c69
--- /dev/null
+++ b/test/CodeGen/AMDGPU/uniform-branch-intrinsic-cond.ll
@@ -0,0 +1,27 @@
+; RUN: llc -march=amdgcn -mcpu=verde -verify-machineinstrs < %s | FileCheck %s
+; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck %s
+
+; This used to raise an assertion due to how the choice between uniform and
+; non-uniform branches was determined.
+;
+; CHECK-LABEL: {{^}}main:
+; CHECK: s_cbranch_vccnz
+define amdgpu_ps float @main(<4 x i32> inreg %rsrc) {
+main_body:
+  %v = call float @llvm.amdgcn.buffer.load.f32(<4 x i32> %rsrc, i32 0, i32 0, i1 true, i1 false)
+  %cc = fcmp une float %v, 1.000000e+00
+  br i1 %cc, label %if, label %else
+
+if:
+  %u = fadd float %v, %v
+  br label %else
+
+else:
+  %r = phi float [ %v, %main_body ], [ %u, %if ]
+  ret float %r
+}
+
+; Function Attrs: nounwind readonly
+declare float @llvm.amdgcn.buffer.load.f32(<4 x i32>, i32, i32, i1, i1) #0
+
+attributes #0 = { nounwind readonly }