[PATCH] D19931: AMDGPU: Uniform branch conditions can originate with intrinsics
Nicolai Hähnle via llvm-commits
llvm-commits at lists.llvm.org
Wed May 4 12:51:56 PDT 2016
nhaehnle created this revision.
nhaehnle added reviewers: tstellarAMD, arsenm.
nhaehnle added a subscriber: llvm-commits.
Herald added a subscriber: arsenm.
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.
http://reviews.llvm.org/D19931
Files:
lib/Target/AMDGPU/SIISelLowering.cpp
test/CodeGen/AMDGPU/uniform-branch-intrinsic-cond.ll
Index: test/CodeGen/AMDGPU/uniform-branch-intrinsic-cond.ll
===================================================================
--- /dev/null
+++ test/CodeGen/AMDGPU/uniform-branch-intrinsic-cond.ll
@@ -0,0 +1,26 @@
+; RUN: llc -march=amdgcn -mcpu=verde -verify-machineinstrs < %s | FileCheck %s
+; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck %s
+
+source_filename = "tgsi"
+target triple = "amdgcn--"
+
+;CHECK-LABEL: {{^}}main:
+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 }
Index: lib/Target/AMDGPU/SIISelLowering.cpp
===================================================================
--- lib/Target/AMDGPU/SIISelLowering.cpp
+++ lib/Target/AMDGPU/SIISelLowering.cpp
@@ -1356,14 +1356,13 @@
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));
-------------- next part --------------
A non-text attachment was scrubbed...
Name: D19931.56189.patch
Type: text/x-patch
Size: 1675 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20160504/cb4c459d/attachment.bin>
More information about the llvm-commits
mailing list