summaryrefslogtreecommitdiffstats
path: root/llvm
diff options
context:
space:
mode:
Diffstat (limited to 'llvm')
-rw-r--r--llvm/lib/Target/AMDGPU/AMDGPUMCInstLower.cpp13
-rw-r--r--llvm/lib/Target/AMDGPU/MCTargetDesc/SIMCCodeEmitter.cpp2
-rw-r--r--llvm/test/CodeGen/AMDGPU/call-encoding.ll19
3 files changed, 30 insertions, 4 deletions
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUMCInstLower.cpp b/llvm/lib/Target/AMDGPU/AMDGPUMCInstLower.cpp
index ba52c3ae1a4..0900832c1f1 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUMCInstLower.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUMCInstLower.cpp
@@ -129,6 +129,7 @@ bool AMDGPUMCInstLower::lowerOperand(const MachineOperand &MO,
void AMDGPUMCInstLower::lower(const MachineInstr *MI, MCInst &OutMI) const {
unsigned Opcode = MI->getOpcode();
+ const auto *TII = ST.getInstrInfo();
// FIXME: Should be able to handle this with emitPseudoExpansionLowering. We
// need to select it to the subtarget specific version, and there's no way to
@@ -137,11 +138,17 @@ void AMDGPUMCInstLower::lower(const MachineInstr *MI, MCInst &OutMI) const {
Opcode = AMDGPU::S_SETPC_B64;
else if (Opcode == AMDGPU::SI_CALL) {
// SI_CALL is just S_SWAPPC_B64 with an additional operand to track the
- // called function.
- Opcode = AMDGPU::S_SWAPPC_B64;
+ // called function (which we need to remove here).
+ OutMI.setOpcode(TII->pseudoToMCOpcode(AMDGPU::S_SWAPPC_B64));
+ MCOperand Dest, Src;
+ lowerOperand(MI->getOperand(0), Dest);
+ lowerOperand(MI->getOperand(1), Src);
+ OutMI.addOperand(Dest);
+ OutMI.addOperand(Src);
+ return;
}
- int MCOpcode = ST.getInstrInfo()->pseudoToMCOpcode(Opcode);
+ int MCOpcode = TII->pseudoToMCOpcode(Opcode);
if (MCOpcode == -1) {
LLVMContext &C = MI->getParent()->getParent()->getFunction()->getContext();
C.emitError("AMDGPUMCInstLower::lower - Pseudo instruction doesn't have "
diff --git a/llvm/lib/Target/AMDGPU/MCTargetDesc/SIMCCodeEmitter.cpp b/llvm/lib/Target/AMDGPU/MCTargetDesc/SIMCCodeEmitter.cpp
index 376c9bfe5cc..94c0157edeb 100644
--- a/llvm/lib/Target/AMDGPU/MCTargetDesc/SIMCCodeEmitter.cpp
+++ b/llvm/lib/Target/AMDGPU/MCTargetDesc/SIMCCodeEmitter.cpp
@@ -278,7 +278,7 @@ void SIMCCodeEmitter::encodeInstruction(const MCInst &MI, raw_ostream &OS,
return;
// Check for additional literals in SRC0/1/2 (Op 1/2/3)
- for (unsigned i = 0, e = MI.getNumOperands(); i < e; ++i) {
+ for (unsigned i = 0, e = Desc.getNumOperands(); i < e; ++i) {
// Check if this operand should be encoded as [SV]Src
if (!AMDGPU::isSISrcOperand(Desc, i))
diff --git a/llvm/test/CodeGen/AMDGPU/call-encoding.ll b/llvm/test/CodeGen/AMDGPU/call-encoding.ll
new file mode 100644
index 00000000000..01e35368de9
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/call-encoding.ll
@@ -0,0 +1,19 @@
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=fiji -amdgpu-function-calls -filetype=obj -verify-machineinstrs < %s | llvm-objdump -triple amdgcn--amdhsa -mcpu=fiji -d - | FileCheck -check-prefixes=GCN,VI %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -amdgpu-function-calls -filetype=obj -verify-machineinstrs < %s | llvm-objdump -triple amdgcn--amdhsa -mcpu=gfx900 -d - | FileCheck -check-prefixes=GCN,GFX9 %s
+; XUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=hawaii -amdgpu-function-calls -filetype=obj -verify-machineinstrs < %s | llvm-objdump -triple amdgcn--amdhsa -mcpu=hawaii -d - | FileCheck -check-prefixes=GCN,CI %s
+
+; GCN: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: s_setpc_b64
+define void @void_func_void() #1 {
+ ret void
+}
+
+; GCN: s_getpc_b64
+; GCN: s_swappc_b64
+define amdgpu_kernel void @test_call_void_func_void() {
+ call void @void_func_void()
+ ret void
+}
+
+attributes #0 = { nounwind }
+attributes #1 = { nounwind noinline }
OpenPOWER on IntegriCloud