6502/llvm-6502
1
0
Fork 0
mirror of https://github.com/c64scene-ar/llvm-6502.git synced 2026-04-20 00:20:11 +00:00
Code Issues Projects Releases Wiki Activity

Actually support volatile memcpys in NVPTX lowering

Browse Source 
Differential Revision: http://reviews.llvm.org/D11091



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@241914 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Eli Bendersky
2015-07-10 15:40:33 +00:00
parent 46c36cfa0f
commit 98da4704dd
2 changed files with 23 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files
lib/Target/NVPTX/NVPTXLowerAggrCopies.cpp
+10 -8
View File
@@ -57,7 +57,6 @@ char NVPTXLowerAggrCopies::ID = 0;
// Lower MemTransferInst or load-store pair to loop
static void convertTransferToLoop(
Instruction *splitAt, Value *srcAddr, Value *dstAddr, Value *len,
//unsigned numLoads,
bool srcVolatile, bool dstVolatile, LLVMContext &Context, Function &F) {
Type *indType = len->getType();
@@ -84,6 +83,8 @@ static void convertTransferToLoop(
ind->addIncoming(ConstantInt::get(indType, 0), origBB);
// load from srcAddr+ind
// TODO: we can leverage the align parameter of llvm.memcpy for more efficient
// word-sized loads and stores.
Value *val = loop.CreateLoad(loop.CreateGEP(loop.getInt8Ty(), srcAddr, ind),
srcVolatile);
// store at dstAddr+ind
@@ -200,13 +201,14 @@ bool NVPTXLowerAggrCopies::runOnFunction(Function &F) {
}
for (MemTransferInst *cpy : aggrMemcpys) {
Value *len = cpy->getLength();
// llvm 2.7 version of memcpy does not have volatile
// operand yet. So always making it non-volatile
// optimistically, so that we don't see unnecessary
// st.volatile in ptx
convertTransferToLoop(cpy, cpy->getSource(), cpy->getDest(), len, false,
false, Context, F);
convertTransferToLoop(/* splitAt */ cpy,
/* srcAddr */ cpy->getSource(),
/* dstAddr */ cpy->getDest(),
/* len */ cpy->getLength(),
/* srcVolatile */ cpy->isVolatile(),
/* dstVolatile */ cpy->isVolatile(),
/* Context */ Context,
/* Function F */ F);
cpy->eraseFromParent();
}
test/CodeGen/NVPTX/lower-aggr-copies.ll
+13
View File
@@ -19,6 +19,19 @@ entry:
; CHECK-NEXT: @%p[[PRED]] bra LBB[[LABEL]]
}
define i8* @memcpy_volatile_caller(i8* %dst, i8* %src, i64 %n) #0 {
entry:
tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %dst, i8* %src, i64 %n, i32 1, i1 true)
ret i8* %dst
; CHECK-LABEL: .visible .func (.param .b32 func_retval0) memcpy_volatile_caller
; CHECK: LBB[[LABEL:[_0-9]+]]:
; CHECK: ld.volatile.u8 %rs[[REG:[0-9]+]]
; CHECK: st.volatile.u8 [%r{{[0-9]+}}], %rs[[REG]]
; CHECK: add.s64 %rd[[COUNTER:[0-9]+]], %rd[[COUNTER]], 1
; CHECK-NEXT: setp.lt.u64 %p[[PRED:[0-9]+]], %rd[[COUNTER]], %rd
; CHECK-NEXT: @%p[[PRED]] bra LBB[[LABEL]]
}
define i8* @memset_caller(i8* %dst, i32 %c, i64 %n) #0 {
entry:
%0 = trunc i32 %c to i8