Skip to content

[NVPTX] Remove 'param' variants of nvvm.ptr.* intrinics #137065

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
15 changes: 2 additions & 13 deletions llvm/include/llvm/IR/IntrinsicsNVVM.td
Original file line number Diff line number Diff line change
Expand Up @@ -116,10 +116,12 @@
// * llvm.nvvm.ptr.gen.to.shared --> ibid.
// * llvm.nvvm.ptr.gen.to.constant --> ibid.
// * llvm.nvvm.ptr.gen.to.local --> ibid.
// * llvm.nvvm.ptr.gen.to.param --> ibid.
// * llvm.nvvm.ptr.global.to.gen --> ibid.
// * llvm.nvvm.ptr.shared.to.gen --> ibid.
// * llvm.nvvm.ptr.constant.to.gen --> ibid.
// * llvm.nvvm.ptr.local.to.gen --> ibid.
// * llvm.nvvm.ptr.param.to.gen --> ibid.
// * llvm.nvvm.ldg.global.i --> load addrspace(1) !load.invariant
// * llvm.nvvm.ldg.global.f --> ibid.
// * llvm.nvvm.ldg.global.p --> ibid.
Expand Down Expand Up @@ -1897,19 +1899,6 @@ def int_nvvm_ldu_global_p : Intrinsic<[llvm_anyptr_ty],
[IntrReadMem, IntrArgMemOnly, IntrNoCallback, IntrWillReturn, NoCapture<ArgIndex<0>>],
"llvm.nvvm.ldu.global.p">;

// Used in nvvm internally to help address space opt and ptx code generation
// This is for params that are passed to kernel functions by pointer by-val.
def int_nvvm_ptr_gen_to_param: Intrinsic<[llvm_anyptr_ty],
[llvm_anyptr_ty],
[IntrNoMem, IntrSpeculatable, IntrNoCallback],
"llvm.nvvm.ptr.gen.to.param">;

// sm70+, PTX7.7+
def int_nvvm_ptr_param_to_gen: DefaultAttrsIntrinsic<[llvm_anyptr_ty],
[llvm_anyptr_ty],
[IntrNoMem, IntrSpeculatable, IntrNoCallback],
"llvm.nvvm.ptr.param.to.gen">;

// Represents an explicit hole in the LLVM IR type system. It may be inserted by
// the compiler in cases where a pointer is of the wrong type. In the backend
// this intrinsic will be folded away and not equate to any instruction. It
Expand Down
26 changes: 12 additions & 14 deletions llvm/lib/IR/AutoUpgrade.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -1054,6 +1054,12 @@ static Intrinsic::ID shouldUpgradeNVPTXBF16Intrinsic(StringRef Name) {
return Intrinsic::not_intrinsic;
}

static bool consumeNVVMPtrAddrSpace(StringRef &Name) {
return Name.consume_front("local") || Name.consume_front("shared") ||
Name.consume_front("global") || Name.consume_front("constant") ||
Name.consume_front("param");
}

static bool upgradeIntrinsicFunction1(Function *F, Function *&NewFn,
bool CanUpgradeDebugIntrinsicsToRecords) {
assert(F && "Illegal to upgrade a non-existent Function.");
Expand Down Expand Up @@ -1361,15 +1367,11 @@ static bool upgradeIntrinsicFunction1(Function *F, Function *&NewFn,
// nvvm.rotate.{b32,b64,right.b64}
Expand = Name == "b32" || Name == "b64" || Name == "right.b64";
else if (Name.consume_front("ptr.gen.to."))
// nvvm.ptr.gen.to.{local,shared,global,constant}
Expand = Name.starts_with("local") || Name.starts_with("shared") ||
Name.starts_with("global") || Name.starts_with("constant");
// nvvm.ptr.gen.to.{local,shared,global,constant,param}
Expand = consumeNVVMPtrAddrSpace(Name);
else if (Name.consume_front("ptr."))
// nvvm.ptr.{local,shared,global,constant}.to.gen
Expand =
(Name.consume_front("local") || Name.consume_front("shared") ||
Name.consume_front("global") || Name.consume_front("constant")) &&
Name.starts_with(".to.gen");
// nvvm.ptr.{local,shared,global,constant,param}.to.gen
Expand = consumeNVVMPtrAddrSpace(Name) && Name.starts_with(".to.gen");
else if (Name.consume_front("ldg.global."))
// nvvm.ldg.global.{i,p,f}
Expand = (Name.starts_with("i.") || Name.starts_with("f.") ||
Expand Down Expand Up @@ -2450,12 +2452,8 @@ static Value *upgradeNVVMIntrinsicCall(StringRef Name, CallBase *CI,
Rep = Builder.CreateIntrinsic(Int64Ty, Intrinsic::fshl,
{Arg, Arg, Builder.getInt64(32)});
} else if ((Name.consume_front("ptr.gen.to.") &&
(Name.starts_with("local") || Name.starts_with("shared") ||
Name.starts_with("global") || Name.starts_with("constant"))) ||
(Name.consume_front("ptr.") &&
(Name.consume_front("local") || Name.consume_front("shared") ||
Name.consume_front("global") ||
Name.consume_front("constant")) &&
consumeNVVMPtrAddrSpace(Name)) ||
(Name.consume_front("ptr.") && consumeNVVMPtrAddrSpace(Name) &&
Name.starts_with(".to.gen"))) {
Rep = Builder.CreateAddrSpaceCast(CI->getArgOperand(0), CI->getType());
} else if (Name.consume_front("ldg.global")) {
Expand Down
12 changes: 0 additions & 12 deletions llvm/lib/Target/NVPTX/NVPTXIntrinsics.td
Original file line number Diff line number Diff line change
Expand Up @@ -2413,18 +2413,6 @@ foreach space = ["local", "shared", "global", "const", "param"] in {
defm cvta_shared_cluster : NG_TO_G<"shared::cluster", false, [hasClusters]>;
defm cvta_to_shared_cluster : G_TO_NG<"shared::cluster", false, [hasClusters]>;

def : Pat<(int_nvvm_ptr_param_to_gen i32:$src),
(cvta_param $src)>;

def : Pat<(int_nvvm_ptr_param_to_gen i64:$src),
(cvta_param_64 $src)>;

// nvvm.ptr.gen.to.param
def : Pat<(int_nvvm_ptr_gen_to_param i32:$src),
(i32 Int32Regs:$src)>;

def : Pat<(int_nvvm_ptr_gen_to_param i64:$src),
(i64 Int64Regs:$src)>;

// nvvm.move intrinsicc
def nvvm_move_i16 : NVPTXInst<(outs Int16Regs:$r), (ins Int16Regs:$s),
Expand Down
7 changes: 7 additions & 0 deletions llvm/test/Assembler/auto_upgrade_nvvm_intrinsics.ll
Original file line number Diff line number Diff line change
Expand Up @@ -44,10 +44,12 @@ declare ptr addrspace(1) @llvm.nvvm.ptr.gen.to.global.p1.p0(ptr)
declare ptr addrspace(3) @llvm.nvvm.ptr.gen.to.shared.p3.p0(ptr)
declare ptr addrspace(4) @llvm.nvvm.ptr.gen.to.constant.p4.p0(ptr)
declare ptr addrspace(5) @llvm.nvvm.ptr.gen.to.local.p5.p0(ptr)
declare ptr addrspace(101) @llvm.nvvm.ptr.gen.to.param.p101.p0(ptr)
declare ptr @llvm.nvvm.ptr.global.to.gen.p0.p1(ptr addrspace(1))
declare ptr @llvm.nvvm.ptr.shared.to.gen.p0.p3(ptr addrspace(3))
declare ptr @llvm.nvvm.ptr.constant.to.gen.p0.p4(ptr addrspace(4))
declare ptr @llvm.nvvm.ptr.local.to.gen.p0.p5(ptr addrspace(5))
declare ptr @llvm.nvvm.ptr.param.to.gen.p0.p101(ptr addrspace(101))

declare i32 @llvm.nvvm.ldg.global.i.i32.p1(ptr addrspace(1), i32)
declare ptr @llvm.nvvm.ldg.global.p.p1(ptr addrspace(1), i32)
Expand Down Expand Up @@ -219,6 +221,8 @@ define void @addrspacecast(ptr %p0) {
; CHECK: %6 = addrspacecast ptr addrspace(4) %5 to ptr
; CHECK: %7 = addrspacecast ptr %6 to ptr addrspace(5)
; CHECK: %8 = addrspacecast ptr addrspace(5) %7 to ptr
; CHECK: %9 = addrspacecast ptr %8 to ptr addrspace(101)
; CHECK: %10 = addrspacecast ptr addrspace(101) %9 to ptr
;
%p1 = call ptr addrspace(1) @llvm.nvvm.ptr.gen.to.global.p1.p0(ptr %p0)
%p2 = call ptr @llvm.nvvm.ptr.global.to.gen.p0.p1(ptr addrspace(1) %p1)
Expand All @@ -232,6 +236,9 @@ define void @addrspacecast(ptr %p0) {
%p7 = call ptr addrspace(5) @llvm.nvvm.ptr.gen.to.local.p5.p0(ptr %p6)
%p8 = call ptr @llvm.nvvm.ptr.local.to.gen.p0.p5(ptr addrspace(5) %p7)

%p9 = call ptr addrspace(101) @llvm.nvvm.ptr.gen.to.param.p101.p0(ptr %p8)
%p10 = call ptr @llvm.nvvm.ptr.param.to.gen.p0.p101(ptr addrspace(101) %p9)

ret void
}

Expand Down
Loading