Skip to content
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.

Sign up for GitHub

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

Update NVVM ldu/ldg intrinsics with IntrWillReturn and test for DCE #98968

Merged
merged 6 commits into from
Jul 24, 2024
Merged

Update NVVM ldu/ldg intrinsics with IntrWillReturn and test for DCE #98968

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
d1b3855
Check for side effects when lowering target intrinsics, update NVVM l…
kalxr Jul 5, 2024
5666ebd
Use function attributes rather than instruction
kalxr Jul 19, 2024
84c3295
No SDAG change
kalxr Jul 22, 2024
a252b0d
Revert ldu-ldg.ll
kalxr Jul 22, 2024
7f8f4f3
Add new test to check that nvvm ldu/ldg intrinsics are DCE'd
kalxr Jul 22, 2024
f66ce52
Update test to run only DCE pass
kalxr Jul 23, 2024
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
12 changes: 6 additions & 6 deletions llvm/include/llvm/IR/IntrinsicsNVVM.td
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1529,30 +1529,30 @@ def int_nvvm_mbarrier_pending_count :
// pointer's alignment.
def int_nvvm_ldu_global_i : Intrinsic<[llvm_anyint_ty],
[llvm_anyptr_ty, llvm_i32_ty],
[IntrReadMem, IntrArgMemOnly, IntrNoCallback, NoCapture<ArgIndex<0>>],
[IntrReadMem, IntrArgMemOnly, IntrNoCallback, IntrWillReturn, NoCapture<ArgIndex<0>>],
"llvm.nvvm.ldu.global.i">;
def int_nvvm_ldu_global_f : Intrinsic<[llvm_anyfloat_ty],
[llvm_anyptr_ty, llvm_i32_ty],
[IntrReadMem, IntrArgMemOnly, IntrNoCallback, NoCapture<ArgIndex<0>>],
[IntrReadMem, IntrArgMemOnly, IntrNoCallback, IntrWillReturn, NoCapture<ArgIndex<0>>],
"llvm.nvvm.ldu.global.f">;
def int_nvvm_ldu_global_p : Intrinsic<[llvm_anyptr_ty],
[llvm_anyptr_ty, llvm_i32_ty],
[IntrReadMem, IntrArgMemOnly, IntrNoCallback, NoCapture<ArgIndex<0>>],
[IntrReadMem, IntrArgMemOnly, IntrNoCallback, IntrWillReturn, NoCapture<ArgIndex<0>>],
"llvm.nvvm.ldu.global.p">;

// Generated within nvvm. Use for ldg on sm_35 or later. Second arg is the
// pointer's alignment.
def int_nvvm_ldg_global_i : Intrinsic<[llvm_anyint_ty],
[llvm_anyptr_ty, llvm_i32_ty],
[IntrReadMem, IntrArgMemOnly, IntrNoCallback, NoCapture<ArgIndex<0>>],
[IntrReadMem, IntrArgMemOnly, IntrNoCallback, IntrWillReturn, NoCapture<ArgIndex<0>>],
"llvm.nvvm.ldg.global.i">;
def int_nvvm_ldg_global_f : Intrinsic<[llvm_anyfloat_ty],
[llvm_anyptr_ty, llvm_i32_ty],
[IntrReadMem, IntrArgMemOnly, IntrNoCallback, NoCapture<ArgIndex<0>>],
[IntrReadMem, IntrArgMemOnly, IntrNoCallback, IntrWillReturn, NoCapture<ArgIndex<0>>],
"llvm.nvvm.ldg.global.f">;
def int_nvvm_ldg_global_p : Intrinsic<[llvm_anyptr_ty],
[llvm_anyptr_ty, llvm_i32_ty],
[IntrReadMem, IntrArgMemOnly, IntrNoCallback, NoCapture<ArgIndex<0>>],
[IntrReadMem, IntrArgMemOnly, IntrNoCallback, IntrWillReturn, NoCapture<ArgIndex<0>>],
"llvm.nvvm.ldg.global.p">;

// Use for generic pointers
Expand Down
187 changes: 187 additions & 0 deletions llvm/test/Transforms/DCE/nvvm-ldu-ldg-willreturn.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,187 @@
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
; RUN: opt -S < %s -passes=dce | FileCheck %s

; ldu/ldg intrinsics were erroneously not marked IntrWillReturn, preventing
; them from being eliminated at IR level when dead.

declare i8 @llvm.nvvm.ldu.global.i.i8.p1(ptr addrspace(1) %ptr, i32 %align)
declare i16 @llvm.nvvm.ldu.global.i.i16.p1(ptr addrspace(1) %ptr, i32 %align)
declare i32 @llvm.nvvm.ldu.global.i.i32.p1(ptr addrspace(1) %ptr, i32 %align)
declare i64 @llvm.nvvm.ldu.global.i.i64.p1(ptr addrspace(1) %ptr, i32 %align)
declare ptr @llvm.nvvm.ldu.global.p.p1(ptr addrspace(1) %ptr, i32 %align)
declare float @llvm.nvvm.ldu.global.f.f32.p1(ptr addrspace(1) %ptr, i32 %align)
declare double @llvm.nvvm.ldu.global.f.f64.p1(ptr addrspace(1) %ptr, i32 %align)
declare half @llvm.nvvm.ldu.global.f.f16.p1(ptr addrspace(1) %ptr, i32 %align)
declare <2 x half> @llvm.nvvm.ldu.global.f.v2f16.p1(ptr addrspace(1) %ptr, i32 %align)

declare i8 @llvm.nvvm.ldg.global.i.i8.p1(ptr addrspace(1) %ptr, i32 %align)
declare i16 @llvm.nvvm.ldg.global.i.i16.p1(ptr addrspace(1) %ptr, i32 %align)
declare i32 @llvm.nvvm.ldg.global.i.i32.p1(ptr addrspace(1) %ptr, i32 %align)
declare i64 @llvm.nvvm.ldg.global.i.i64.p1(ptr addrspace(1) %ptr, i32 %align)
declare ptr @llvm.nvvm.ldg.global.p.p1(ptr addrspace(1) %ptr, i32 %align)
declare float @llvm.nvvm.ldg.global.f.f32.p1(ptr addrspace(1) %ptr, i32 %align)
declare double @llvm.nvvm.ldg.global.f.f64.p1(ptr addrspace(1) %ptr, i32 %align)
declare half @llvm.nvvm.ldg.global.f.f16.p1(ptr addrspace(1) %ptr, i32 %align)
declare <2 x half> @llvm.nvvm.ldg.global.f.v2f16.p1(ptr addrspace(1) %ptr, i32 %align)

define void @test_ldu_i8_dead(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define void @test_ldu_i8_dead(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) {
; CHECK-NEXT: ret void
;
%val = tail call i8 @llvm.nvvm.ldu.global.i.i8.p1(ptr addrspace(1) %ptr, i32 4)
ret void
}

define void @test_ldu_i16_dead(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define void @test_ldu_i16_dead(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) {
; CHECK-NEXT: ret void
;
%val = tail call i16 @llvm.nvvm.ldu.global.i.i16.p1(ptr addrspace(1) %ptr, i32 2)
ret void
}

define void @test_ldu_i32_dead(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define void @test_ldu_i32_dead(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) {
; CHECK-NEXT: ret void
;
%val = tail call i32 @llvm.nvvm.ldu.global.i.i32.p1(ptr addrspace(1) %ptr, i32 4)
ret void
}

define void @test_ldu_i64_dead(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define void @test_ldu_i64_dead(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) {
; CHECK-NEXT: ret void
;
%val = tail call i64 @llvm.nvvm.ldu.global.i.i64.p1(ptr addrspace(1) %ptr, i32 8)
ret void
}

define void @test_ldu_p_dead(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define void @test_ldu_p_dead(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) {
; CHECK-NEXT: ret void
;
%val = tail call ptr @llvm.nvvm.ldu.global.p.p1(ptr addrspace(1) %ptr, i32 8)
ret void
}

define void @test_ldu_f32_dead(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define void @test_ldu_f32_dead(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) {
; CHECK-NEXT: ret void
;
%val = tail call float @llvm.nvvm.ldu.global.f.f32.p1(ptr addrspace(1) %ptr, i32 4)
ret void
}

define void @test_ldu_f64_dead(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define void @test_ldu_f64_dead(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) {
; CHECK-NEXT: ret void
;
%val = tail call double @llvm.nvvm.ldu.global.f.f64.p1(ptr addrspace(1) %ptr, i32 8)
ret void
}

define void @test_ldu_f16_dead(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define void @test_ldu_f16_dead(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) {
; CHECK-NEXT: ret void
;
%val = tail call half @llvm.nvvm.ldu.global.f.f16.p1(ptr addrspace(1) %ptr, i32 2)
ret void
}

define void @test_ldu_v2f16_dead(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define void @test_ldu_v2f16_dead(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) {
; CHECK-NEXT: ret void
;
%val = tail call <2 x half> @llvm.nvvm.ldu.global.f.v2f16.p1(ptr addrspace(1) %ptr, i32 4)
ret void
}

define void @test_ldg_i8_dead(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define void @test_ldg_i8_dead(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) {
; CHECK-NEXT: ret void
;
%val = tail call i8 @llvm.nvvm.ldg.global.i.i8.p1(ptr addrspace(1) %ptr, i32 4)
ret void
}

define void @test_ldg_i16_dead(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define void @test_ldg_i16_dead(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) {
; CHECK-NEXT: ret void
;
%val = tail call i16 @llvm.nvvm.ldg.global.i.i16.p1(ptr addrspace(1) %ptr, i32 2)
ret void
}

define void @test_ldg_i32_dead(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define void @test_ldg_i32_dead(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) {
; CHECK-NEXT: ret void
;
%val = tail call i32 @llvm.nvvm.ldg.global.i.i32.p1(ptr addrspace(1) %ptr, i32 4)
ret void
}

define void @test_ldg_i64_dead(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define void @test_ldg_i64_dead(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) {
; CHECK-NEXT: ret void
;
%val = tail call i64 @llvm.nvvm.ldg.global.i.i64.p1(ptr addrspace(1) %ptr, i32 8)
ret void
}

define void @test_ldg_p_dead(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define void @test_ldg_p_dead(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) {
; CHECK-NEXT: ret void
;
%val = tail call ptr @llvm.nvvm.ldg.global.p.p1(ptr addrspace(1) %ptr, i32 8)
ret void
}

define void @test_ldg_f32_dead(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define void @test_ldg_f32_dead(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) {
; CHECK-NEXT: ret void
;
%val = tail call float @llvm.nvvm.ldg.global.f.f32.p1(ptr addrspace(1) %ptr, i32 4)
ret void
}

define void @test_ldg_f64_dead(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define void @test_ldg_f64_dead(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) {
; CHECK-NEXT: ret void
;
%val = tail call double @llvm.nvvm.ldg.global.f.f64.p1(ptr addrspace(1) %ptr, i32 8)
ret void
}

define void @test_ldg_f16_dead(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define void @test_ldg_f16_dead(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) {
; CHECK-NEXT: ret void
;
%val = tail call half @llvm.nvvm.ldg.global.f.f16.p1(ptr addrspace(1) %ptr, i32 2)
ret void
}

define void @test_ldg_v2f16_dead(ptr addrspace(1) %ptr) {
; CHECK-LABEL: define void @test_ldg_v2f16_dead(
; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) {
; CHECK-NEXT: ret void
;
%val = tail call <2 x half> @llvm.nvvm.ldg.global.f.v2f16.p1(ptr addrspace(1) %ptr, i32 4)
ret void
}
Loading