[Flang][mlir] - Translation of delayed privatization for deferred target-tasks #155348
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
bhandarkar-pranav
requested review from
Meinersbur,
TIFitis,
agozillon,
ergawy,
jsjodin,
skatrak and
tblah
llvmbot
added
mlir:core
Member
|
@llvm/pr-subscribers-mlir-llvm @llvm/pr-subscribers-flang-driver Author: Pranav Bhandarkar (bhandarkar-pranav) ChangesThis PR adds support for translation of the private clause on deferred target tasks - that is An offloading call for a deferred target-task is not blocking - the offloading host task continues it execution after issuing the offloading call. Therefore, the key problem we need to solve is to ensure that the data needed for private variables to be initialized in the target task persists even after the host task has completed.
The pass uses a rewrite pattern applied using the greedy pattern matcher, which in turn does some constant folding and DCE. Due to this a number of lit tests had to be updated. In GEPs constant get folded into indices and truncated to i32 types. In some tests sequence of insertvalue and extractvalue instructions get cancelled out. So, these needed to be updated too. Patch is 79.33 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/155348.diff 30 Files Affected:
diff --git a/flang/include/flang/Optimizer/Passes/Pipelines.h b/flang/include/flang/Optimizer/Passes/Pipelines.h
index a3f59ee8dd013..17d48f46e4b9b 100644
--- a/flang/include/flang/Optimizer/Passes/Pipelines.h
+++ b/flang/include/flang/Optimizer/Passes/Pipelines.h
@@ -22,6 +22,7 @@
#include "mlir/Conversion/SCFToControlFlow/SCFToControlFlow.h"
#include "mlir/Dialect/GPU/IR/GPUDialect.h"
#include "mlir/Dialect/LLVMIR/LLVMAttrs.h"
+#include "mlir/Dialect/LLVMIR/Transforms/OpenMPOffloadPrivatizationPrepare.h"
#include "mlir/Pass/PassManager.h"
#include "mlir/Transforms/GreedyPatternRewriteDriver.h"
#include "mlir/Transforms/Passes.h"
diff --git a/flang/lib/Optimizer/Passes/Pipelines.cpp b/flang/lib/Optimizer/Passes/Pipelines.cpp
index ca8e820608688..6a11461cd8380 100644
--- a/flang/lib/Optimizer/Passes/Pipelines.cpp
+++ b/flang/lib/Optimizer/Passes/Pipelines.cpp
@@ -403,6 +403,12 @@ void createMLIRToLLVMPassPipeline(mlir::PassManager &pm,
// Add codegen pass pipeline.
fir::createDefaultFIRCodeGenPassPipeline(pm, config, inputFilename);
+
+ // Run a pass to prepare for translation of delayed privatization in the
+ // context of deferred target tasks.
+ addNestedPassConditionally<mlir::LLVM::LLVMFuncOp>(pm, disableFirToLlvmIr,[&]() {
+ return mlir::LLVM::createPrepareForOMPOffloadPrivatizationPass();
+ });
}
} // namespace fir
diff --git a/flang/test/Driver/tco-emit-final-mlir.fir b/flang/test/Driver/tco-emit-final-mlir.fir
index 75f8f153127af..177810cf41378 100644
--- a/flang/test/Driver/tco-emit-final-mlir.fir
+++ b/flang/test/Driver/tco-emit-final-mlir.fir
@@ -13,7 +13,7 @@
// CHECK: llvm.return
// CHECK-NOT: func.func
-func.func @_QPfoo() {
+func.func @_QPfoo() -> !fir.ref<i32> {
%1 = fir.alloca i32
- return
+ return %1 : !fir.ref<i32>
}
diff --git a/flang/test/Driver/tco-test-gen.fir b/flang/test/Driver/tco-test-gen.fir
index 38d4e50ecf3aa..15483f7ee3534 100644
--- a/flang/test/Driver/tco-test-gen.fir
+++ b/flang/test/Driver/tco-test-gen.fir
@@ -42,11 +42,10 @@ func.func @_QPtest(%arg0: !fir.ref<i32> {fir.bindc_name = "num"}, %arg1: !fir.re
// CHECK-SAME: %[[ARG2:.*]]: !llvm.ptr {fir.bindc_name = "ub", llvm.nocapture},
// CHECK-SAME: %[[ARG3:.*]]: !llvm.ptr {fir.bindc_name = "step", llvm.nocapture}) {
+// CMPLX: %[[VAL_3:.*]] = llvm.mlir.constant(0 : index) : i64
+// CMPLX: %[[VAL_2:.*]] = llvm.mlir.constant(1 : index) : i64
// CMPLX: %[[VAL_0:.*]] = llvm.mlir.constant(1 : i64) : i64
// CMPLX: %[[VAL_1:.*]] = llvm.alloca %[[VAL_0]] x i32 {bindc_name = "i"} : (i64) -> !llvm.ptr
-// CMPLX: %[[VAL_2:.*]] = llvm.mlir.constant(1 : index) : i64
-// CMPLX: %[[VAL_3:.*]] = llvm.mlir.constant(0 : index) : i64
-// CMPLX: %[[VAL_4:.*]] = llvm.mlir.constant(1 : i64) : i64
// SIMPLE: %[[VAL_3:.*]] = llvm.mlir.constant(0 : index) : i64
// SIMPLE: %[[VAL_2:.*]] = llvm.mlir.constant(1 : index) : i64
diff --git a/flang/test/Fir/alloc-32.fir b/flang/test/Fir/alloc-32.fir
index a3cbf200c24fc..f57f6ce6fcf5e 100644
--- a/flang/test/Fir/alloc-32.fir
+++ b/flang/test/Fir/alloc-32.fir
@@ -19,7 +19,7 @@ func.func @allocmem_scalar_nonchar() -> !fir.heap<i32> {
// CHECK-LABEL: define ptr @allocmem_scalar_dynchar(
// CHECK-SAME: i32 %[[len:.*]])
// CHECK: %[[mul1:.*]] = sext i32 %[[len]] to i64
-// CHECK: %[[mul2:.*]] = mul i64 1, %[[mul1]]
+// CHECK: %[[mul2:.*]] = mul i64 %[[mul1]], 1
// CHECK: %[[cmp:.*]] = icmp sgt i64 %[[mul2]], 0
// CHECK: %[[sz:.*]] = select i1 %[[cmp]], i64 %[[mul2]], i64 1
// CHECK: %[[trunc:.*]] = trunc i64 %[[sz]] to i32
diff --git a/flang/test/Fir/alloc.fir b/flang/test/Fir/alloc.fir
index 8da8b828c18b9..0d3ce323d0d7c 100644
--- a/flang/test/Fir/alloc.fir
+++ b/flang/test/Fir/alloc.fir
@@ -86,7 +86,7 @@ func.func @alloca_scalar_dynchar_kind(%l : i32) -> !fir.ref<!fir.char<2,?>> {
// CHECK-LABEL: define ptr @allocmem_scalar_dynchar(
// CHECK-SAME: i32 %[[len:.*]])
// CHECK: %[[mul1:.*]] = sext i32 %[[len]] to i64
-// CHECK: %[[mul2:.*]] = mul i64 1, %[[mul1]]
+// CHECK: %[[mul2:.*]] = mul i64 %[[mul1]], 1
// CHECK: %[[cmp:.*]] = icmp sgt i64 %[[mul2]], 0
// CHECK: %[[size:.*]] = select i1 %[[cmp]], i64 %[[mul2]], i64 1
// CHECK: call ptr @malloc(i64 %[[size]])
@@ -98,7 +98,7 @@ func.func @allocmem_scalar_dynchar(%l : i32) -> !fir.heap<!fir.char<1,?>> {
// CHECK-LABEL: define ptr @allocmem_scalar_dynchar_kind(
// CHECK-SAME: i32 %[[len:.*]])
// CHECK: %[[mul1:.*]] = sext i32 %[[len]] to i64
-// CHECK: %[[mul2:.*]] = mul i64 2, %[[mul1]]
+// CHECK: %[[mul2:.*]] = mul i64 %[[mul1]], 2
// CHECK: %[[cmp:.*]] = icmp sgt i64 %[[mul2]], 0
// CHECK: %[[size:.*]] = select i1 %[[cmp]], i64 %[[mul2]], i64 1
// CHECK: call ptr @malloc(i64 %[[size]])
@@ -185,7 +185,7 @@ func.func @alloca_dynarray_of_nonchar2(%e: index) -> !fir.ref<!fir.array<?x?xi32
// CHECK-LABEL: define ptr @allocmem_dynarray_of_nonchar(
// CHECK-SAME: i64 %[[extent:.*]])
-// CHECK: %[[prod1:.*]] = mul i64 12, %[[extent]]
+// CHECK: %[[prod1:.*]] = mul i64 %[[extent]], 12
// CHECK: %[[cmp:.*]] = icmp sgt i64 %[[prod1]], 0
// CHECK: %[[size:.*]] = select i1 %[[cmp]], i64 %[[prod1]], i64 1
// CHECK: call ptr @malloc(i64 %[[size]])
@@ -196,7 +196,7 @@ func.func @allocmem_dynarray_of_nonchar(%e: index) -> !fir.heap<!fir.array<3x?xi
// CHECK-LABEL: define ptr @allocmem_dynarray_of_nonchar2(
// CHECK-SAME: i64 %[[extent:.*]])
-// CHECK: %[[prod1:.*]] = mul i64 4, %[[extent]]
+// CHECK: %[[prod1:.*]] = mul i64 %[[extent]], 4
// CHECK: %[[prod2:.*]] = mul i64 %[[prod1]], %[[extent]]
// CHECK: %[[cmp:.*]] = icmp sgt i64 %[[prod2]], 0
// CHECK: %[[size:.*]] = select i1 %[[cmp]], i64 %[[prod2]], i64 1
@@ -227,7 +227,7 @@ func.func @alloca_dynarray_of_char2(%e : index) -> !fir.ref<!fir.array<?x?x!fir.
// CHECK-LABEL: define ptr @allocmem_dynarray_of_char(
// CHECK-SAME: i64 %[[extent:.*]])
-// CHECK: %[[prod1:.*]] = mul i64 60, %[[extent]]
+// CHECK: %[[prod1:.*]] = mul i64 %[[extent]], 60
// CHECK: %[[cmp:.*]] = icmp sgt i64 %[[prod1]], 0
// CHECK: %[[size:.*]] = select i1 %[[cmp]], i64 %[[prod1]], i64 1
// CHECK: call ptr @malloc(i64 %[[size]])
@@ -238,7 +238,7 @@ func.func @allocmem_dynarray_of_char(%e : index) -> !fir.heap<!fir.array<3x?x!fi
// CHECK-LABEL: define ptr @allocmem_dynarray_of_char2(
// CHECK-SAME: i64 %[[extent:.*]])
-// CHECK: %[[prod1:.*]] = mul i64 20, %[[extent]]
+// CHECK: %[[prod1:.*]] = mul i64 %[[extent]], 20
// CHECK: %[[prod2:.*]] = mul i64 %[[prod1]], %[[extent]]
// CHECK: %[[cmp:.*]] = icmp sgt i64 %[[prod2]], 0
// CHECK: %[[size:.*]] = select i1 %[[cmp]], i64 %[[mul2]], i64 1
@@ -286,7 +286,7 @@ func.func @allocmem_dynarray_of_dynchar(%l: i32, %e : index) -> !fir.heap<!fir.a
// CHECK-LABEL: define ptr @allocmem_dynarray_of_dynchar2(
// CHECK-SAME: i32 %[[len:.*]], i64 %[[extent:.*]])
// CHECK: %[[a:.*]] = sext i32 %[[len]] to i64
-// CHECK: %[[prod1:.*]] = mul i64 2, %[[a]]
+// CHECK: %[[prod1:.*]] = mul i64 %[[a]], 2
// CHECK: %[[prod2:.*]] = mul i64 %[[prod1]], %[[extent]]
// CHECK: %[[prod3:.*]] = mul i64 %[[prod2]], %[[extent]]
// CHECK: %[[cmp:.*]] = icmp sgt i64 %[[prod3]], 0
@@ -366,12 +366,13 @@ func.func @allocmem_array_with_holes_dynchar(%arg0: index, %arg1: index) -> !fir
// CHECK: %[[VAL_0:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, ptr, [1 x i64] }, i64 1
// CHECK: %[[VAL_3:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]], ptr, [1 x i64] }, i64 1
// CHECK: %[[VAL_2:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, ptr, [1 x i64] }, i64 1
-
+func.func private @foo(%0: !fir.ref<!fir.class<none>>, %1: !fir.ref<!fir.class<!fir.array<?xnone>>>, %2: !fir.ref<!fir.box<none>>, %3: !fir.ref<!fir.box<!fir.array<?xnone>>>)
func.func @alloca_unlimited_polymorphic_box() {
%0 = fir.alloca !fir.class<none>
%1 = fir.alloca !fir.class<!fir.array<?xnone>>
%2 = fir.alloca !fir.box<none>
%3 = fir.alloca !fir.box<!fir.array<?xnone>>
+ fir.call @foo(%0, %1, %2, %3) : (!fir.ref<!fir.class<none>>, !fir.ref<!fir.class<!fir.array<?xnone>>>, !fir.ref<!fir.box<none>>, !fir.ref<!fir.box<!fir.array<?xnone>>>) -> ()
return
}
// Note: allocmem of fir.box are not possible (fir::HeapType::verify does not
diff --git a/flang/test/Fir/arrexp.fir b/flang/test/Fir/arrexp.fir
index e8ec8ac79e0c2..2eb717228d998 100644
--- a/flang/test/Fir/arrexp.fir
+++ b/flang/test/Fir/arrexp.fir
@@ -143,9 +143,9 @@ func.func @f6(%arg0: !fir.box<!fir.array<?xf32>>, %arg1: f32) {
%c9 = arith.constant 9 : index
%c10 = arith.constant 10 : index
- // CHECK: %[[EXT_GEP:.*]] = getelementptr {{.*}} %[[A]], i32 0, i32 7, i64 0, i32 1
+ // CHECK: %[[EXT_GEP:.*]] = getelementptr {{.*}} %[[A]], i32 0, i32 7, i32 0, i32 1
// CHECK: %[[EXTENT:.*]] = load i64, ptr %[[EXT_GEP]]
- // CHECK: %[[SIZE:.*]] = mul i64 4, %[[EXTENT]]
+ // CHECK: %[[SIZE:.*]] = mul i64 %[[EXTENT]], 4
// CHECK: %[[CMP:.*]] = icmp sgt i64 %[[SIZE]], 0
// CHECK: %[[SZ:.*]] = select i1 %[[CMP]], i64 %[[SIZE]], i64 1
// CHECK: %[[MALLOC:.*]] = call ptr @malloc(i64 %[[SZ]])
diff --git a/flang/test/Fir/basic-program.fir b/flang/test/Fir/basic-program.fir
index c9fe53bf093a1..6bad03dded24d 100644
--- a/flang/test/Fir/basic-program.fir
+++ b/flang/test/Fir/basic-program.fir
@@ -158,4 +158,6 @@ func.func @_QQmain() {
// PASSES-NEXT: LowerNontemporalPass
// PASSES-NEXT: FIRToLLVMLowering
// PASSES-NEXT: ReconcileUnrealizedCasts
+// PASSES-NEXT: 'llvm.func' Pipeline
+// PASSES-NEXT: PrepareForOMPOffloadPrivatizationPass
// PASSES-NEXT: LLVMIRLoweringPass
diff --git a/flang/test/Fir/box.fir b/flang/test/Fir/box.fir
index c0cf3d8375983..760fbd4792122 100644
--- a/flang/test/Fir/box.fir
+++ b/flang/test/Fir/box.fir
@@ -57,7 +57,7 @@ func.func @fa(%a : !fir.ref<!fir.array<100xf32>>) {
// CHECK-SAME: ptr {{[^%]*}}%[[res:.*]], ptr {{[^%]*}}%[[arg0:.*]], i64 %[[arg1:.*]])
func.func @b1(%arg0 : !fir.ref<!fir.char<1,?>>, %arg1 : index) -> !fir.box<!fir.char<1,?>> {
// CHECK: %[[alloca:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8 }
- // CHECK: %[[size:.*]] = mul i64 1, %[[arg1]]
+ // CHECK: %[[size:.*]] = mul i64 %[[arg1]], 1
// CHECK: insertvalue {{.*}} undef, i64 %[[size]], 1
// CHECK: insertvalue {{.*}} i32 20240719, 2
// CHECK: insertvalue {{.*}} ptr %[[arg0]], 0
@@ -89,7 +89,7 @@ func.func @b2(%arg0 : !fir.ref<!fir.array<?x!fir.char<1,5>>>, %arg1 : index) ->
func.func @b3(%arg0 : !fir.ref<!fir.array<?x!fir.char<1,?>>>, %arg1 : index, %arg2 : index) -> !fir.box<!fir.array<?x!fir.char<1,?>>> {
%1 = fir.shape %arg2 : (index) -> !fir.shape<1>
// CHECK: %[[alloca:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }
- // CHECK: %[[size:.*]] = mul i64 1, %[[arg1]]
+ // CHECK: %[[size:.*]] = mul i64 %[[arg1]], 1
// CHECK: insertvalue {{.*}} i64 %[[size]], 1
// CHECK: insertvalue {{.*}} i32 20240719, 2
// CHECK: insertvalue {{.*}} i64 %[[arg2]], 7, 0, 1
@@ -108,7 +108,7 @@ func.func @b4(%arg0 : !fir.ref<!fir.array<7x!fir.char<1,?>>>, %arg1 : index) ->
%c_7 = arith.constant 7 : index
%1 = fir.shape %c_7 : (index) -> !fir.shape<1>
// CHECK: %[[alloca:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }
- // CHECK: %[[size:.*]] = mul i64 1, %[[arg1]]
+ // CHECK: %[[size:.*]] = mul i64 %[[arg1]], 1
// CHECK: insertvalue {{.*}} i64 %[[size]], 1
// CHECK: insertvalue {{.*}} i32 20240719, 2
// CHECK: insertvalue {{.*}} i64 7, 7, 0, 1
diff --git a/flang/test/Fir/boxproc.fir b/flang/test/Fir/boxproc.fir
index 97d9b38ed6f40..d4c36a4f5b213 100644
--- a/flang/test/Fir/boxproc.fir
+++ b/flang/test/Fir/boxproc.fir
@@ -82,12 +82,8 @@ func.func @_QPtest_proc_dummy_other(%arg0: !fir.boxproc<() -> ()>) {
// CHECK: store [1 x i8] c" ", ptr %[[VAL_18]], align 1
// CHECK: call void @llvm.init.trampoline(ptr %[[VAL_20]], ptr @_QFtest_proc_dummy_charPgen_message, ptr %[[VAL_2]])
// CHECK: %[[VAL_23:.*]] = call ptr @llvm.adjust.trampoline(ptr %[[VAL_20]])
-// CHECK: %[[VAL_25:.*]] = insertvalue { ptr, i64 } undef, ptr %[[VAL_23]], 0
-// CHECK: %[[VAL_26:.*]] = insertvalue { ptr, i64 } %[[VAL_25]], i64 10, 1
// CHECK: %[[VAL_27:.*]] = call ptr @llvm.stacksave.p0()
-// CHECK: %[[VAL_28:.*]] = extractvalue { ptr, i64 } %[[VAL_26]], 0
-// CHECK: %[[VAL_29:.*]] = extractvalue { ptr, i64 } %[[VAL_26]], 1
-// CHECK: %[[VAL_30:.*]] = call { ptr, i64 } @_QPget_message(ptr %[[VAL_0]], i64 40, ptr %[[VAL_28]], i64 %[[VAL_29]])
+// CHECK: %[[VAL_30:.*]] = call { ptr, i64 } @_QPget_message(ptr %[[VAL_0]], i64 40, ptr %[[VAL_23]], i64 10)
// CHECK: %[[VAL_32:.*]] = call i1 @_FortranAioOutputAscii(ptr %{{.*}}, ptr %[[VAL_0]], i64 40)
// CHECK: call void @llvm.stackrestore.p0(ptr %[[VAL_27]])
@@ -115,14 +111,10 @@ func.func @_QPtest_proc_dummy_other(%arg0: !fir.boxproc<() -> ()>) {
// CHECK-LABEL: define { ptr, i64 } @_QPget_message(ptr
// CHECK-SAME: %[[VAL_0:.*]], i64 %[[VAL_1:.*]], ptr %[[VAL_2:.*]], i64
// CHECK-SAME: %[[VAL_3:.*]])
-// CHECK: %[[VAL_4:.*]] = insertvalue { ptr, i64 } undef, ptr %[[VAL_2]], 0
-// CHECK: %[[VAL_5:.*]] = insertvalue { ptr, i64 } %[[VAL_4]], i64 %[[VAL_3]], 1
-// CHECK: %[[VAL_7:.*]] = extractvalue { ptr, i64 } %[[VAL_5]], 0
-// CHECK: %[[VAL_8:.*]] = extractvalue { ptr, i64 } %[[VAL_5]], 1
// CHECK: %[[VAL_9:.*]] = call ptr @llvm.stacksave.p0()
-// CHECK: %[[VAL_10:.*]] = alloca i8, i64 %[[VAL_8]], align 1
-// CHECK: %[[VAL_12:.*]] = call { ptr, i64 } %[[VAL_7]](ptr %[[VAL_10]], i64 %[[VAL_8]])
-// CHECK: %[[VAL_13:.*]] = add i64 %[[VAL_8]], 12
+// CHECK: %[[VAL_10:.*]] = alloca i8, i64 %[[VAL_3]], align 1
+// CHECK: %[[VAL_12:.*]] = call { ptr, i64 } %[[VAL_2]](ptr %[[VAL_10]], i64 %[[VAL_3]])
+// CHECK: %[[VAL_13:.*]] = add i64 %[[VAL_3]], 12
// CHECK: %[[VAL_14:.*]] = alloca i8, i64 %[[VAL_13]], align 1
// CHECK: call void @llvm.memmove.p0.p0.i64(ptr %[[VAL_14]], ptr {{.*}}, i64 12, i1 false)
// CHECK: %[[VAL_18:.*]] = phi i64
diff --git a/flang/test/Fir/embox.fir b/flang/test/Fir/embox.fir
index 0f304cff2c79e..11f7457b6873c 100644
--- a/flang/test/Fir/embox.fir
+++ b/flang/test/Fir/embox.fir
@@ -11,7 +11,7 @@ func.func @_QPtest_callee(%arg0: !fir.box<!fir.array<?xi32>>) {
func.func @_QPtest_slice() {
// CHECK: %[[a1:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }, align 8
// CHECK: %[[a2:.*]] = alloca [20 x i32], i64 1, align 4
-// CHECK: %[[a3:.*]] = getelementptr [20 x i32], ptr %[[a2]], i64 0, i64 0
+// CHECK: %[[a3:.*]] = getelementptr [20 x i32], ptr %[[a2]], i32 0, i64 0
// CHECK: %[[a4:.*]] = insertvalue { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }
// CHECK: { ptr undef, i64 4, i32 20240719, i8 1, i8 9, i8 0, i8 0, [1 x [3 x i64]]
// CHECK: [i64 1, i64 5, i64 8]] }, ptr %[[a3]], 0
@@ -38,7 +38,7 @@ func.func @_QPtest_dt_callee(%arg0: !fir.box<!fir.array<?xi32>>) {
func.func @_QPtest_dt_slice() {
// CHECK: %[[a1:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }, align 8
// CHECK: %[[a3:.*]] = alloca [20 x %_QFtest_dt_sliceTt], i64 1, align 8
-// CHECK: %[[a4:.*]] = getelementptr [20 x %_QFtest_dt_sliceTt], ptr %[[a3]], i64 0, i64 0, i32 0
+// CHECK: %[[a4:.*]] = getelementptr [20 x %_QFtest_dt_sliceTt], ptr %[[a3]], i32 0, i64 0, i32 0
// CHECK: %[[a5:.*]] = insertvalue { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }
// CHECK-SAME: { ptr undef, i64 4, i32 20240719, i8 1, i8 9, i8 0, i8 0, [1 x [3 x i64]]
// CHECK-SAME: [i64 1, i64 5, i64 16
@@ -73,7 +73,7 @@ func.func @emboxSubstring(%arg0: !fir.ref<!fir.array<2x3x!fir.char<1,4>>>) {
%0 = fir.shape %c2, %c3 : (index, index) -> !fir.shape<2>
%1 = fir.slice %c1, %c2, %c1, %c1, %c3, %c1 substr %c1_i64, %c2_i64 : (index, index, index, index, index, index, i64, i64) -> !fir.slice<2>
%2 = fir.embox %arg0(%0) [%1] : (!fir.ref<!fir.array<2x3x!fir.char<1,4>>>, !fir.shape<2>, !fir.slice<2>) -> !fir.box<!fir.array<?x?x!fir.char<1,?>>>
- // CHECK: %[[addr:.*]] = getelementptr [3 x [2 x [4 x i8]]], ptr %[[arg0]], i64 0, i64 0, i64 0, i64 1
+ // CHECK: %[[addr:.*]] = getelementptr [3 x [2 x [4 x i8]]], ptr %[[arg0]], i32 0, i64 0, i64 0, i32 1
// CHECK: insertvalue {[[descriptorType:.*]]} { ptr undef, i64 2, i32 20240719, i8 2, i8 40, i8 0, i8 0
// CHECK-SAME: [2 x [3 x i64]] [{{\[}}3 x i64] [i64 1, i64 2, i64 4], [3 x i64] [i64 1, i64 3, i64 8]] }
// CHECK-SAME: ptr %[[addr]], 0
diff --git a/flang/test/Fir/omp-reduction-embox-codegen.fir b/flang/test/Fir/omp-reduction-embox-codegen.fir
index 1645e1a407ad4..e517b1352ff5c 100644
--- a/flang/test/Fir/omp-reduction-embox-codegen.fir
+++ b/flang/test/Fir/omp-reduction-embox-codegen.fir
@@ -23,14 +23,14 @@ omp.declare_reduction @test_reduction : !fir.ref<!fir.box<i32>> init {
omp.yield(%0 : !fir.ref<!fir.box<i32>>)
}
-func.func @_QQmain() attributes {fir.bindc_name = "reduce"} {
+func.func @_QQmain() -> !fir.ref<!fir.box<i32>> attributes {fir.bindc_name = "reduce"} {
%4 = fir.alloca !fir.box<i32>
omp.parallel reduction(byref @test_reduction %4 -> %arg0 : !fir.ref<!fir.box<i32>>) {
omp.terminator
}
- return
+ return %4: !fir.ref<!fir.box<i32>>
}
// basically we are testing that there isn't a crash
-// CHECK-LABEL: define void @_QQmain
+// CHECK-LABEL: define ptr @_QQmain
// CHECK-NEXT: alloca { ptr, i64, i32, i8, i8, i8, i8 }, i64 1, align 8
diff --git a/flang/test/Fir/optional.fir b/flang/test/Fir/optional.fir
index bded8b5332a30..66ff69f083467 100644
--- a/flang/test/Fir/optional.fir
+++ b/flang/test/Fir/optional.fir
@@ -37,8 +37,7 @@ func.func @bar2() -> i1 {
// CHECK-LABEL: @foo3
func.func @foo3(%arg0: !fir.boxchar<1>) -> i1 {
- // CHECK: %[[extract:.*]] = extractvalue { ptr, i64 } %{{.*}}, 0
- // CHECK: %[[ptr:.*]] = ptrtoint ptr %[[extract]] to i64
+ // CHECK: %[[ptr:.*]] = ptrtoint ptr %0 to i64
// CHECK: icmp ne i64 %[[ptr]], 0
%0 = fir.is_present %arg0 : (!fir.boxchar<1>) -> i1
return %0 : i1
diff --git a/flang/test/Fir/pdt.fir b/flang/test/Fir/pdt.fir
index a200cd7e7cc03..411927aae6bdf 100644
--- a/flang/test/Fir/pdt.fir
+++ b/flang/test/Fir/pdt.fir
@@ -96,13 +96,13 @@ func.func @_QTt1P.f2.offset(%0 : i32, %1 : i32) -> i32 {
func.func private @bar(!fir.ref<!fir.char<1,?>>)
-// CHECK-LABEL: define void @_QPfoo(i32 %0, i32 %1)
-func.func @_QPfoo(%arg0 : i32, %arg1 : i32) {
+// CHECK-LABEL: define ptr @_QPfoo(i32 %0, i32 %1)
+func.func @_QPfoo(%arg0 : i32, %arg1 : i32) -> !fir.ref<!fir.type<_QTt1>> {
// CHECK: %[[size:.*]] = call i64 @_QTt1P.mem.size(i32 %0, i32 %1)
// CHECK: %[[alloc:.*]] = alloca i8, i64 %[[size]]
%0 = fir.alloca !fir.type<_QTt1(p1:i32,p2:i32){f1:!fir.char<1,?>,f2:!fir.char<1,?>}>(%arg0, %arg1 : i32, i32)
//%2 = fir.coordinate_of %0, f2 : (!fir.ref<!fir.type<_QTt1>>) -> !fir.ref<!fir.char<1,?>>
%2 = fir.zero_bits !fir.ref<!fir.char<1,?>>
fir.call @bar(%2) : (!fir.ref<!fir.char<1,?>>) -> ()
- return
+ return %0 : !fir.ref<!fir.type<_QTt1>>
}
diff --git a/flang/test/Fir/rebox.fir b/flang/test/Fir/rebox.fir
index 0c9f6d9bb94ad..d858adfb7c45d 100644
--- a/flang/test/Fir/rebox.fir
+++ b/flang/test/Fir/rebox.fir
@@ -36,7 +36,7 @@ func.func @test_rebox_1(%arg0: !fir.box<!fir.array<?x?xf32>>) {
// CHECK: %[[VOIDBASE0:.*]] = getelementptr i8, ptr %[[INBASE]], i64 %[[OFFSET_0]]
// CHECK: %[[OFFSET_1:.*]] = mul i64 2, %[[INSTRIDE_1]]
// CHECK: %[[VOIDBASE1:.*]] = getelementptr i8, ptr %[[VOIDBASE0]], i64 %[[OFFSET_1]]
- // CHECK: %[[OUTSTRIDE0:.*]] = mul i64 3, %[[INSTRIDE_1]]
+ // CHECK: %[[OUTSTRIDE0:.*]] = mul i64 %[[INSTRIDE_1]], 3
// CHECK: %[[OUTBOX1:.*]] = insertvalue { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] } %{{.*}}, i64 %[[OUTSTRIDE0]], 7, 0, 2
// CHECK: %[[OUTBOX2:.*]] = insertvalue { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] } %[[OUTBOX1]], ptr %[[VOIDBASE1]], 0
// CHECK: store { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] } %[[OUTBOX2]], ptr %[[OUTBOX_ALLOC]], align 8
@@ -63,7 +63,7 @@ func.func @test_rebox_2(%arg0: !fir.box<!fir.array<?x?x!fir.char<1,?>>>) {
// CHECK: %[[OUTBOX:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, [2 x [3 x i64]] }
// CHECK: %[[LEN_GEP:.*]] = getelementptr { ptr, i64, i32, i8, i8, i8, i8, [2 x [3 x i64]] }, ptr %[[INBOX]], i32 0, i32 1
// CHECK: %[[LEN:...
[truncated]
|
Member
|
@llvm/pr-subscribers-mlir-openmp Author: Pranav Bhandarkar (bhandarkar-pranav) ChangesThis PR adds support for translation of the private clause on deferred target tasks - that is An offloading call for a deferred target-task is not blocking - the offloading host task continues it execution after issuing the offloading call. Therefore, the key problem we need to solve is to ensure that the data needed for private variables to be initialized in the target task persists even after the host task has completed.
The pass uses a rewrite pattern applied using the greedy pattern matcher, which in turn does some constant folding and DCE. Due to this a number of lit tests had to be updated. In GEPs constant get folded into indices and truncated to i32 types. In some tests sequence of insertvalue and extractvalue instructions get cancelled out. So, these needed to be updated too. Patch is 79.33 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/155348.diff 30 Files Affected:
diff --git a/flang/include/flang/Optimizer/Passes/Pipelines.h b/flang/include/flang/Optimizer/Passes/Pipelines.h
index a3f59ee8dd013..17d48f46e4b9b 100644
--- a/flang/include/flang/Optimizer/Passes/Pipelines.h
+++ b/flang/include/flang/Optimizer/Passes/Pipelines.h
@@ -22,6 +22,7 @@
#include "mlir/Conversion/SCFToControlFlow/SCFToControlFlow.h"
#include "mlir/Dialect/GPU/IR/GPUDialect.h"
#include "mlir/Dialect/LLVMIR/LLVMAttrs.h"
+#include "mlir/Dialect/LLVMIR/Transforms/OpenMPOffloadPrivatizationPrepare.h"
#include "mlir/Pass/PassManager.h"
#include "mlir/Transforms/GreedyPatternRewriteDriver.h"
#include "mlir/Transforms/Passes.h"
diff --git a/flang/lib/Optimizer/Passes/Pipelines.cpp b/flang/lib/Optimizer/Passes/Pipelines.cpp
index ca8e820608688..6a11461cd8380 100644
--- a/flang/lib/Optimizer/Passes/Pipelines.cpp
+++ b/flang/lib/Optimizer/Passes/Pipelines.cpp
@@ -403,6 +403,12 @@ void createMLIRToLLVMPassPipeline(mlir::PassManager &pm,
// Add codegen pass pipeline.
fir::createDefaultFIRCodeGenPassPipeline(pm, config, inputFilename);
+
+ // Run a pass to prepare for translation of delayed privatization in the
+ // context of deferred target tasks.
+ addNestedPassConditionally<mlir::LLVM::LLVMFuncOp>(pm, disableFirToLlvmIr,[&]() {
+ return mlir::LLVM::createPrepareForOMPOffloadPrivatizationPass();
+ });
}
} // namespace fir
diff --git a/flang/test/Driver/tco-emit-final-mlir.fir b/flang/test/Driver/tco-emit-final-mlir.fir
index 75f8f153127af..177810cf41378 100644
--- a/flang/test/Driver/tco-emit-final-mlir.fir
+++ b/flang/test/Driver/tco-emit-final-mlir.fir
@@ -13,7 +13,7 @@
// CHECK: llvm.return
// CHECK-NOT: func.func
-func.func @_QPfoo() {
+func.func @_QPfoo() -> !fir.ref<i32> {
%1 = fir.alloca i32
- return
+ return %1 : !fir.ref<i32>
}
diff --git a/flang/test/Driver/tco-test-gen.fir b/flang/test/Driver/tco-test-gen.fir
index 38d4e50ecf3aa..15483f7ee3534 100644
--- a/flang/test/Driver/tco-test-gen.fir
+++ b/flang/test/Driver/tco-test-gen.fir
@@ -42,11 +42,10 @@ func.func @_QPtest(%arg0: !fir.ref<i32> {fir.bindc_name = "num"}, %arg1: !fir.re
// CHECK-SAME: %[[ARG2:.*]]: !llvm.ptr {fir.bindc_name = "ub", llvm.nocapture},
// CHECK-SAME: %[[ARG3:.*]]: !llvm.ptr {fir.bindc_name = "step", llvm.nocapture}) {
+// CMPLX: %[[VAL_3:.*]] = llvm.mlir.constant(0 : index) : i64
+// CMPLX: %[[VAL_2:.*]] = llvm.mlir.constant(1 : index) : i64
// CMPLX: %[[VAL_0:.*]] = llvm.mlir.constant(1 : i64) : i64
// CMPLX: %[[VAL_1:.*]] = llvm.alloca %[[VAL_0]] x i32 {bindc_name = "i"} : (i64) -> !llvm.ptr
-// CMPLX: %[[VAL_2:.*]] = llvm.mlir.constant(1 : index) : i64
-// CMPLX: %[[VAL_3:.*]] = llvm.mlir.constant(0 : index) : i64
-// CMPLX: %[[VAL_4:.*]] = llvm.mlir.constant(1 : i64) : i64
// SIMPLE: %[[VAL_3:.*]] = llvm.mlir.constant(0 : index) : i64
// SIMPLE: %[[VAL_2:.*]] = llvm.mlir.constant(1 : index) : i64
diff --git a/flang/test/Fir/alloc-32.fir b/flang/test/Fir/alloc-32.fir
index a3cbf200c24fc..f57f6ce6fcf5e 100644
--- a/flang/test/Fir/alloc-32.fir
+++ b/flang/test/Fir/alloc-32.fir
@@ -19,7 +19,7 @@ func.func @allocmem_scalar_nonchar() -> !fir.heap<i32> {
// CHECK-LABEL: define ptr @allocmem_scalar_dynchar(
// CHECK-SAME: i32 %[[len:.*]])
// CHECK: %[[mul1:.*]] = sext i32 %[[len]] to i64
-// CHECK: %[[mul2:.*]] = mul i64 1, %[[mul1]]
+// CHECK: %[[mul2:.*]] = mul i64 %[[mul1]], 1
// CHECK: %[[cmp:.*]] = icmp sgt i64 %[[mul2]], 0
// CHECK: %[[sz:.*]] = select i1 %[[cmp]], i64 %[[mul2]], i64 1
// CHECK: %[[trunc:.*]] = trunc i64 %[[sz]] to i32
diff --git a/flang/test/Fir/alloc.fir b/flang/test/Fir/alloc.fir
index 8da8b828c18b9..0d3ce323d0d7c 100644
--- a/flang/test/Fir/alloc.fir
+++ b/flang/test/Fir/alloc.fir
@@ -86,7 +86,7 @@ func.func @alloca_scalar_dynchar_kind(%l : i32) -> !fir.ref<!fir.char<2,?>> {
// CHECK-LABEL: define ptr @allocmem_scalar_dynchar(
// CHECK-SAME: i32 %[[len:.*]])
// CHECK: %[[mul1:.*]] = sext i32 %[[len]] to i64
-// CHECK: %[[mul2:.*]] = mul i64 1, %[[mul1]]
+// CHECK: %[[mul2:.*]] = mul i64 %[[mul1]], 1
// CHECK: %[[cmp:.*]] = icmp sgt i64 %[[mul2]], 0
// CHECK: %[[size:.*]] = select i1 %[[cmp]], i64 %[[mul2]], i64 1
// CHECK: call ptr @malloc(i64 %[[size]])
@@ -98,7 +98,7 @@ func.func @allocmem_scalar_dynchar(%l : i32) -> !fir.heap<!fir.char<1,?>> {
// CHECK-LABEL: define ptr @allocmem_scalar_dynchar_kind(
// CHECK-SAME: i32 %[[len:.*]])
// CHECK: %[[mul1:.*]] = sext i32 %[[len]] to i64
-// CHECK: %[[mul2:.*]] = mul i64 2, %[[mul1]]
+// CHECK: %[[mul2:.*]] = mul i64 %[[mul1]], 2
// CHECK: %[[cmp:.*]] = icmp sgt i64 %[[mul2]], 0
// CHECK: %[[size:.*]] = select i1 %[[cmp]], i64 %[[mul2]], i64 1
// CHECK: call ptr @malloc(i64 %[[size]])
@@ -185,7 +185,7 @@ func.func @alloca_dynarray_of_nonchar2(%e: index) -> !fir.ref<!fir.array<?x?xi32
// CHECK-LABEL: define ptr @allocmem_dynarray_of_nonchar(
// CHECK-SAME: i64 %[[extent:.*]])
-// CHECK: %[[prod1:.*]] = mul i64 12, %[[extent]]
+// CHECK: %[[prod1:.*]] = mul i64 %[[extent]], 12
// CHECK: %[[cmp:.*]] = icmp sgt i64 %[[prod1]], 0
// CHECK: %[[size:.*]] = select i1 %[[cmp]], i64 %[[prod1]], i64 1
// CHECK: call ptr @malloc(i64 %[[size]])
@@ -196,7 +196,7 @@ func.func @allocmem_dynarray_of_nonchar(%e: index) -> !fir.heap<!fir.array<3x?xi
// CHECK-LABEL: define ptr @allocmem_dynarray_of_nonchar2(
// CHECK-SAME: i64 %[[extent:.*]])
-// CHECK: %[[prod1:.*]] = mul i64 4, %[[extent]]
+// CHECK: %[[prod1:.*]] = mul i64 %[[extent]], 4
// CHECK: %[[prod2:.*]] = mul i64 %[[prod1]], %[[extent]]
// CHECK: %[[cmp:.*]] = icmp sgt i64 %[[prod2]], 0
// CHECK: %[[size:.*]] = select i1 %[[cmp]], i64 %[[prod2]], i64 1
@@ -227,7 +227,7 @@ func.func @alloca_dynarray_of_char2(%e : index) -> !fir.ref<!fir.array<?x?x!fir.
// CHECK-LABEL: define ptr @allocmem_dynarray_of_char(
// CHECK-SAME: i64 %[[extent:.*]])
-// CHECK: %[[prod1:.*]] = mul i64 60, %[[extent]]
+// CHECK: %[[prod1:.*]] = mul i64 %[[extent]], 60
// CHECK: %[[cmp:.*]] = icmp sgt i64 %[[prod1]], 0
// CHECK: %[[size:.*]] = select i1 %[[cmp]], i64 %[[prod1]], i64 1
// CHECK: call ptr @malloc(i64 %[[size]])
@@ -238,7 +238,7 @@ func.func @allocmem_dynarray_of_char(%e : index) -> !fir.heap<!fir.array<3x?x!fi
// CHECK-LABEL: define ptr @allocmem_dynarray_of_char2(
// CHECK-SAME: i64 %[[extent:.*]])
-// CHECK: %[[prod1:.*]] = mul i64 20, %[[extent]]
+// CHECK: %[[prod1:.*]] = mul i64 %[[extent]], 20
// CHECK: %[[prod2:.*]] = mul i64 %[[prod1]], %[[extent]]
// CHECK: %[[cmp:.*]] = icmp sgt i64 %[[prod2]], 0
// CHECK: %[[size:.*]] = select i1 %[[cmp]], i64 %[[mul2]], i64 1
@@ -286,7 +286,7 @@ func.func @allocmem_dynarray_of_dynchar(%l: i32, %e : index) -> !fir.heap<!fir.a
// CHECK-LABEL: define ptr @allocmem_dynarray_of_dynchar2(
// CHECK-SAME: i32 %[[len:.*]], i64 %[[extent:.*]])
// CHECK: %[[a:.*]] = sext i32 %[[len]] to i64
-// CHECK: %[[prod1:.*]] = mul i64 2, %[[a]]
+// CHECK: %[[prod1:.*]] = mul i64 %[[a]], 2
// CHECK: %[[prod2:.*]] = mul i64 %[[prod1]], %[[extent]]
// CHECK: %[[prod3:.*]] = mul i64 %[[prod2]], %[[extent]]
// CHECK: %[[cmp:.*]] = icmp sgt i64 %[[prod3]], 0
@@ -366,12 +366,13 @@ func.func @allocmem_array_with_holes_dynchar(%arg0: index, %arg1: index) -> !fir
// CHECK: %[[VAL_0:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, ptr, [1 x i64] }, i64 1
// CHECK: %[[VAL_3:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]], ptr, [1 x i64] }, i64 1
// CHECK: %[[VAL_2:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, ptr, [1 x i64] }, i64 1
-
+func.func private @foo(%0: !fir.ref<!fir.class<none>>, %1: !fir.ref<!fir.class<!fir.array<?xnone>>>, %2: !fir.ref<!fir.box<none>>, %3: !fir.ref<!fir.box<!fir.array<?xnone>>>)
func.func @alloca_unlimited_polymorphic_box() {
%0 = fir.alloca !fir.class<none>
%1 = fir.alloca !fir.class<!fir.array<?xnone>>
%2 = fir.alloca !fir.box<none>
%3 = fir.alloca !fir.box<!fir.array<?xnone>>
+ fir.call @foo(%0, %1, %2, %3) : (!fir.ref<!fir.class<none>>, !fir.ref<!fir.class<!fir.array<?xnone>>>, !fir.ref<!fir.box<none>>, !fir.ref<!fir.box<!fir.array<?xnone>>>) -> ()
return
}
// Note: allocmem of fir.box are not possible (fir::HeapType::verify does not
diff --git a/flang/test/Fir/arrexp.fir b/flang/test/Fir/arrexp.fir
index e8ec8ac79e0c2..2eb717228d998 100644
--- a/flang/test/Fir/arrexp.fir
+++ b/flang/test/Fir/arrexp.fir
@@ -143,9 +143,9 @@ func.func @f6(%arg0: !fir.box<!fir.array<?xf32>>, %arg1: f32) {
%c9 = arith.constant 9 : index
%c10 = arith.constant 10 : index
- // CHECK: %[[EXT_GEP:.*]] = getelementptr {{.*}} %[[A]], i32 0, i32 7, i64 0, i32 1
+ // CHECK: %[[EXT_GEP:.*]] = getelementptr {{.*}} %[[A]], i32 0, i32 7, i32 0, i32 1
// CHECK: %[[EXTENT:.*]] = load i64, ptr %[[EXT_GEP]]
- // CHECK: %[[SIZE:.*]] = mul i64 4, %[[EXTENT]]
+ // CHECK: %[[SIZE:.*]] = mul i64 %[[EXTENT]], 4
// CHECK: %[[CMP:.*]] = icmp sgt i64 %[[SIZE]], 0
// CHECK: %[[SZ:.*]] = select i1 %[[CMP]], i64 %[[SIZE]], i64 1
// CHECK: %[[MALLOC:.*]] = call ptr @malloc(i64 %[[SZ]])
diff --git a/flang/test/Fir/basic-program.fir b/flang/test/Fir/basic-program.fir
index c9fe53bf093a1..6bad03dded24d 100644
--- a/flang/test/Fir/basic-program.fir
+++ b/flang/test/Fir/basic-program.fir
@@ -158,4 +158,6 @@ func.func @_QQmain() {
// PASSES-NEXT: LowerNontemporalPass
// PASSES-NEXT: FIRToLLVMLowering
// PASSES-NEXT: ReconcileUnrealizedCasts
+// PASSES-NEXT: 'llvm.func' Pipeline
+// PASSES-NEXT: PrepareForOMPOffloadPrivatizationPass
// PASSES-NEXT: LLVMIRLoweringPass
diff --git a/flang/test/Fir/box.fir b/flang/test/Fir/box.fir
index c0cf3d8375983..760fbd4792122 100644
--- a/flang/test/Fir/box.fir
+++ b/flang/test/Fir/box.fir
@@ -57,7 +57,7 @@ func.func @fa(%a : !fir.ref<!fir.array<100xf32>>) {
// CHECK-SAME: ptr {{[^%]*}}%[[res:.*]], ptr {{[^%]*}}%[[arg0:.*]], i64 %[[arg1:.*]])
func.func @b1(%arg0 : !fir.ref<!fir.char<1,?>>, %arg1 : index) -> !fir.box<!fir.char<1,?>> {
// CHECK: %[[alloca:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8 }
- // CHECK: %[[size:.*]] = mul i64 1, %[[arg1]]
+ // CHECK: %[[size:.*]] = mul i64 %[[arg1]], 1
// CHECK: insertvalue {{.*}} undef, i64 %[[size]], 1
// CHECK: insertvalue {{.*}} i32 20240719, 2
// CHECK: insertvalue {{.*}} ptr %[[arg0]], 0
@@ -89,7 +89,7 @@ func.func @b2(%arg0 : !fir.ref<!fir.array<?x!fir.char<1,5>>>, %arg1 : index) ->
func.func @b3(%arg0 : !fir.ref<!fir.array<?x!fir.char<1,?>>>, %arg1 : index, %arg2 : index) -> !fir.box<!fir.array<?x!fir.char<1,?>>> {
%1 = fir.shape %arg2 : (index) -> !fir.shape<1>
// CHECK: %[[alloca:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }
- // CHECK: %[[size:.*]] = mul i64 1, %[[arg1]]
+ // CHECK: %[[size:.*]] = mul i64 %[[arg1]], 1
// CHECK: insertvalue {{.*}} i64 %[[size]], 1
// CHECK: insertvalue {{.*}} i32 20240719, 2
// CHECK: insertvalue {{.*}} i64 %[[arg2]], 7, 0, 1
@@ -108,7 +108,7 @@ func.func @b4(%arg0 : !fir.ref<!fir.array<7x!fir.char<1,?>>>, %arg1 : index) ->
%c_7 = arith.constant 7 : index
%1 = fir.shape %c_7 : (index) -> !fir.shape<1>
// CHECK: %[[alloca:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }
- // CHECK: %[[size:.*]] = mul i64 1, %[[arg1]]
+ // CHECK: %[[size:.*]] = mul i64 %[[arg1]], 1
// CHECK: insertvalue {{.*}} i64 %[[size]], 1
// CHECK: insertvalue {{.*}} i32 20240719, 2
// CHECK: insertvalue {{.*}} i64 7, 7, 0, 1
diff --git a/flang/test/Fir/boxproc.fir b/flang/test/Fir/boxproc.fir
index 97d9b38ed6f40..d4c36a4f5b213 100644
--- a/flang/test/Fir/boxproc.fir
+++ b/flang/test/Fir/boxproc.fir
@@ -82,12 +82,8 @@ func.func @_QPtest_proc_dummy_other(%arg0: !fir.boxproc<() -> ()>) {
// CHECK: store [1 x i8] c" ", ptr %[[VAL_18]], align 1
// CHECK: call void @llvm.init.trampoline(ptr %[[VAL_20]], ptr @_QFtest_proc_dummy_charPgen_message, ptr %[[VAL_2]])
// CHECK: %[[VAL_23:.*]] = call ptr @llvm.adjust.trampoline(ptr %[[VAL_20]])
-// CHECK: %[[VAL_25:.*]] = insertvalue { ptr, i64 } undef, ptr %[[VAL_23]], 0
-// CHECK: %[[VAL_26:.*]] = insertvalue { ptr, i64 } %[[VAL_25]], i64 10, 1
// CHECK: %[[VAL_27:.*]] = call ptr @llvm.stacksave.p0()
-// CHECK: %[[VAL_28:.*]] = extractvalue { ptr, i64 } %[[VAL_26]], 0
-// CHECK: %[[VAL_29:.*]] = extractvalue { ptr, i64 } %[[VAL_26]], 1
-// CHECK: %[[VAL_30:.*]] = call { ptr, i64 } @_QPget_message(ptr %[[VAL_0]], i64 40, ptr %[[VAL_28]], i64 %[[VAL_29]])
+// CHECK: %[[VAL_30:.*]] = call { ptr, i64 } @_QPget_message(ptr %[[VAL_0]], i64 40, ptr %[[VAL_23]], i64 10)
// CHECK: %[[VAL_32:.*]] = call i1 @_FortranAioOutputAscii(ptr %{{.*}}, ptr %[[VAL_0]], i64 40)
// CHECK: call void @llvm.stackrestore.p0(ptr %[[VAL_27]])
@@ -115,14 +111,10 @@ func.func @_QPtest_proc_dummy_other(%arg0: !fir.boxproc<() -> ()>) {
// CHECK-LABEL: define { ptr, i64 } @_QPget_message(ptr
// CHECK-SAME: %[[VAL_0:.*]], i64 %[[VAL_1:.*]], ptr %[[VAL_2:.*]], i64
// CHECK-SAME: %[[VAL_3:.*]])
-// CHECK: %[[VAL_4:.*]] = insertvalue { ptr, i64 } undef, ptr %[[VAL_2]], 0
-// CHECK: %[[VAL_5:.*]] = insertvalue { ptr, i64 } %[[VAL_4]], i64 %[[VAL_3]], 1
-// CHECK: %[[VAL_7:.*]] = extractvalue { ptr, i64 } %[[VAL_5]], 0
-// CHECK: %[[VAL_8:.*]] = extractvalue { ptr, i64 } %[[VAL_5]], 1
// CHECK: %[[VAL_9:.*]] = call ptr @llvm.stacksave.p0()
-// CHECK: %[[VAL_10:.*]] = alloca i8, i64 %[[VAL_8]], align 1
-// CHECK: %[[VAL_12:.*]] = call { ptr, i64 } %[[VAL_7]](ptr %[[VAL_10]], i64 %[[VAL_8]])
-// CHECK: %[[VAL_13:.*]] = add i64 %[[VAL_8]], 12
+// CHECK: %[[VAL_10:.*]] = alloca i8, i64 %[[VAL_3]], align 1
+// CHECK: %[[VAL_12:.*]] = call { ptr, i64 } %[[VAL_2]](ptr %[[VAL_10]], i64 %[[VAL_3]])
+// CHECK: %[[VAL_13:.*]] = add i64 %[[VAL_3]], 12
// CHECK: %[[VAL_14:.*]] = alloca i8, i64 %[[VAL_13]], align 1
// CHECK: call void @llvm.memmove.p0.p0.i64(ptr %[[VAL_14]], ptr {{.*}}, i64 12, i1 false)
// CHECK: %[[VAL_18:.*]] = phi i64
diff --git a/flang/test/Fir/embox.fir b/flang/test/Fir/embox.fir
index 0f304cff2c79e..11f7457b6873c 100644
--- a/flang/test/Fir/embox.fir
+++ b/flang/test/Fir/embox.fir
@@ -11,7 +11,7 @@ func.func @_QPtest_callee(%arg0: !fir.box<!fir.array<?xi32>>) {
func.func @_QPtest_slice() {
// CHECK: %[[a1:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }, align 8
// CHECK: %[[a2:.*]] = alloca [20 x i32], i64 1, align 4
-// CHECK: %[[a3:.*]] = getelementptr [20 x i32], ptr %[[a2]], i64 0, i64 0
+// CHECK: %[[a3:.*]] = getelementptr [20 x i32], ptr %[[a2]], i32 0, i64 0
// CHECK: %[[a4:.*]] = insertvalue { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }
// CHECK: { ptr undef, i64 4, i32 20240719, i8 1, i8 9, i8 0, i8 0, [1 x [3 x i64]]
// CHECK: [i64 1, i64 5, i64 8]] }, ptr %[[a3]], 0
@@ -38,7 +38,7 @@ func.func @_QPtest_dt_callee(%arg0: !fir.box<!fir.array<?xi32>>) {
func.func @_QPtest_dt_slice() {
// CHECK: %[[a1:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }, align 8
// CHECK: %[[a3:.*]] = alloca [20 x %_QFtest_dt_sliceTt], i64 1, align 8
-// CHECK: %[[a4:.*]] = getelementptr [20 x %_QFtest_dt_sliceTt], ptr %[[a3]], i64 0, i64 0, i32 0
+// CHECK: %[[a4:.*]] = getelementptr [20 x %_QFtest_dt_sliceTt], ptr %[[a3]], i32 0, i64 0, i32 0
// CHECK: %[[a5:.*]] = insertvalue { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }
// CHECK-SAME: { ptr undef, i64 4, i32 20240719, i8 1, i8 9, i8 0, i8 0, [1 x [3 x i64]]
// CHECK-SAME: [i64 1, i64 5, i64 16
@@ -73,7 +73,7 @@ func.func @emboxSubstring(%arg0: !fir.ref<!fir.array<2x3x!fir.char<1,4>>>) {
%0 = fir.shape %c2, %c3 : (index, index) -> !fir.shape<2>
%1 = fir.slice %c1, %c2, %c1, %c1, %c3, %c1 substr %c1_i64, %c2_i64 : (index, index, index, index, index, index, i64, i64) -> !fir.slice<2>
%2 = fir.embox %arg0(%0) [%1] : (!fir.ref<!fir.array<2x3x!fir.char<1,4>>>, !fir.shape<2>, !fir.slice<2>) -> !fir.box<!fir.array<?x?x!fir.char<1,?>>>
- // CHECK: %[[addr:.*]] = getelementptr [3 x [2 x [4 x i8]]], ptr %[[arg0]], i64 0, i64 0, i64 0, i64 1
+ // CHECK: %[[addr:.*]] = getelementptr [3 x [2 x [4 x i8]]], ptr %[[arg0]], i32 0, i64 0, i64 0, i32 1
// CHECK: insertvalue {[[descriptorType:.*]]} { ptr undef, i64 2, i32 20240719, i8 2, i8 40, i8 0, i8 0
// CHECK-SAME: [2 x [3 x i64]] [{{\[}}3 x i64] [i64 1, i64 2, i64 4], [3 x i64] [i64 1, i64 3, i64 8]] }
// CHECK-SAME: ptr %[[addr]], 0
diff --git a/flang/test/Fir/omp-reduction-embox-codegen.fir b/flang/test/Fir/omp-reduction-embox-codegen.fir
index 1645e1a407ad4..e517b1352ff5c 100644
--- a/flang/test/Fir/omp-reduction-embox-codegen.fir
+++ b/flang/test/Fir/omp-reduction-embox-codegen.fir
@@ -23,14 +23,14 @@ omp.declare_reduction @test_reduction : !fir.ref<!fir.box<i32>> init {
omp.yield(%0 : !fir.ref<!fir.box<i32>>)
}
-func.func @_QQmain() attributes {fir.bindc_name = "reduce"} {
+func.func @_QQmain() -> !fir.ref<!fir.box<i32>> attributes {fir.bindc_name = "reduce"} {
%4 = fir.alloca !fir.box<i32>
omp.parallel reduction(byref @test_reduction %4 -> %arg0 : !fir.ref<!fir.box<i32>>) {
omp.terminator
}
- return
+ return %4: !fir.ref<!fir.box<i32>>
}
// basically we are testing that there isn't a crash
-// CHECK-LABEL: define void @_QQmain
+// CHECK-LABEL: define ptr @_QQmain
// CHECK-NEXT: alloca { ptr, i64, i32, i8, i8, i8, i8 }, i64 1, align 8
diff --git a/flang/test/Fir/optional.fir b/flang/test/Fir/optional.fir
index bded8b5332a30..66ff69f083467 100644
--- a/flang/test/Fir/optional.fir
+++ b/flang/test/Fir/optional.fir
@@ -37,8 +37,7 @@ func.func @bar2() -> i1 {
// CHECK-LABEL: @foo3
func.func @foo3(%arg0: !fir.boxchar<1>) -> i1 {
- // CHECK: %[[extract:.*]] = extractvalue { ptr, i64 } %{{.*}}, 0
- // CHECK: %[[ptr:.*]] = ptrtoint ptr %[[extract]] to i64
+ // CHECK: %[[ptr:.*]] = ptrtoint ptr %0 to i64
// CHECK: icmp ne i64 %[[ptr]], 0
%0 = fir.is_present %arg0 : (!fir.boxchar<1>) -> i1
return %0 : i1
diff --git a/flang/test/Fir/pdt.fir b/flang/test/Fir/pdt.fir
index a200cd7e7cc03..411927aae6bdf 100644
--- a/flang/test/Fir/pdt.fir
+++ b/flang/test/Fir/pdt.fir
@@ -96,13 +96,13 @@ func.func @_QTt1P.f2.offset(%0 : i32, %1 : i32) -> i32 {
func.func private @bar(!fir.ref<!fir.char<1,?>>)
-// CHECK-LABEL: define void @_QPfoo(i32 %0, i32 %1)
-func.func @_QPfoo(%arg0 : i32, %arg1 : i32) {
+// CHECK-LABEL: define ptr @_QPfoo(i32 %0, i32 %1)
+func.func @_QPfoo(%arg0 : i32, %arg1 : i32) -> !fir.ref<!fir.type<_QTt1>> {
// CHECK: %[[size:.*]] = call i64 @_QTt1P.mem.size(i32 %0, i32 %1)
// CHECK: %[[alloc:.*]] = alloca i8, i64 %[[size]]
%0 = fir.alloca !fir.type<_QTt1(p1:i32,p2:i32){f1:!fir.char<1,?>,f2:!fir.char<1,?>}>(%arg0, %arg1 : i32, i32)
//%2 = fir.coordinate_of %0, f2 : (!fir.ref<!fir.type<_QTt1>>) -> !fir.ref<!fir.char<1,?>>
%2 = fir.zero_bits !fir.ref<!fir.char<1,?>>
fir.call @bar(%2) : (!fir.ref<!fir.char<1,?>>) -> ()
- return
+ return %0 : !fir.ref<!fir.type<_QTt1>>
}
diff --git a/flang/test/Fir/rebox.fir b/flang/test/Fir/rebox.fir
index 0c9f6d9bb94ad..d858adfb7c45d 100644
--- a/flang/test/Fir/rebox.fir
+++ b/flang/test/Fir/rebox.fir
@@ -36,7 +36,7 @@ func.func @test_rebox_1(%arg0: !fir.box<!fir.array<?x?xf32>>) {
// CHECK: %[[VOIDBASE0:.*]] = getelementptr i8, ptr %[[INBASE]], i64 %[[OFFSET_0]]
// CHECK: %[[OFFSET_1:.*]] = mul i64 2, %[[INSTRIDE_1]]
// CHECK: %[[VOIDBASE1:.*]] = getelementptr i8, ptr %[[VOIDBASE0]], i64 %[[OFFSET_1]]
- // CHECK: %[[OUTSTRIDE0:.*]] = mul i64 3, %[[INSTRIDE_1]]
+ // CHECK: %[[OUTSTRIDE0:.*]] = mul i64 %[[INSTRIDE_1]], 3
// CHECK: %[[OUTBOX1:.*]] = insertvalue { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] } %{{.*}}, i64 %[[OUTSTRIDE0]], 7, 0, 2
// CHECK: %[[OUTBOX2:.*]] = insertvalue { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] } %[[OUTBOX1]], ptr %[[VOIDBASE1]], 0
// CHECK: store { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] } %[[OUTBOX2]], ptr %[[OUTBOX_ALLOC]], align 8
@@ -63,7 +63,7 @@ func.func @test_rebox_2(%arg0: !fir.box<!fir.array<?x?x!fir.char<1,?>>>) {
// CHECK: %[[OUTBOX:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, [2 x [3 x i64]] }
// CHECK: %[[LEN_GEP:.*]] = getelementptr { ptr, i64, i32, i8, i8, i8, i8, [2 x [3 x i64]] }, ptr %[[INBOX]], i32 0, i32 1
// CHECK: %[[LEN:...
[truncated]
|
Member
|
@llvm/pr-subscribers-flang-openmp Author: Pranav Bhandarkar (bhandarkar-pranav) ChangesThis PR adds support for translation of the private clause on deferred target tasks - that is An offloading call for a deferred target-task is not blocking - the offloading host task continues it execution after issuing the offloading call. Therefore, the key problem we need to solve is to ensure that the data needed for private variables to be initialized in the target task persists even after the host task has completed.
The pass uses a rewrite pattern applied using the greedy pattern matcher, which in turn does some constant folding and DCE. Due to this a number of lit tests had to be updated. In GEPs constant get folded into indices and truncated to i32 types. In some tests sequence of insertvalue and extractvalue instructions get cancelled out. So, these needed to be updated too. Patch is 79.33 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/155348.diff 30 Files Affected:
diff --git a/flang/include/flang/Optimizer/Passes/Pipelines.h b/flang/include/flang/Optimizer/Passes/Pipelines.h
index a3f59ee8dd013..17d48f46e4b9b 100644
--- a/flang/include/flang/Optimizer/Passes/Pipelines.h
+++ b/flang/include/flang/Optimizer/Passes/Pipelines.h
@@ -22,6 +22,7 @@
#include "mlir/Conversion/SCFToControlFlow/SCFToControlFlow.h"
#include "mlir/Dialect/GPU/IR/GPUDialect.h"
#include "mlir/Dialect/LLVMIR/LLVMAttrs.h"
+#include "mlir/Dialect/LLVMIR/Transforms/OpenMPOffloadPrivatizationPrepare.h"
#include "mlir/Pass/PassManager.h"
#include "mlir/Transforms/GreedyPatternRewriteDriver.h"
#include "mlir/Transforms/Passes.h"
diff --git a/flang/lib/Optimizer/Passes/Pipelines.cpp b/flang/lib/Optimizer/Passes/Pipelines.cpp
index ca8e820608688..6a11461cd8380 100644
--- a/flang/lib/Optimizer/Passes/Pipelines.cpp
+++ b/flang/lib/Optimizer/Passes/Pipelines.cpp
@@ -403,6 +403,12 @@ void createMLIRToLLVMPassPipeline(mlir::PassManager &pm,
// Add codegen pass pipeline.
fir::createDefaultFIRCodeGenPassPipeline(pm, config, inputFilename);
+
+ // Run a pass to prepare for translation of delayed privatization in the
+ // context of deferred target tasks.
+ addNestedPassConditionally<mlir::LLVM::LLVMFuncOp>(pm, disableFirToLlvmIr,[&]() {
+ return mlir::LLVM::createPrepareForOMPOffloadPrivatizationPass();
+ });
}
} // namespace fir
diff --git a/flang/test/Driver/tco-emit-final-mlir.fir b/flang/test/Driver/tco-emit-final-mlir.fir
index 75f8f153127af..177810cf41378 100644
--- a/flang/test/Driver/tco-emit-final-mlir.fir
+++ b/flang/test/Driver/tco-emit-final-mlir.fir
@@ -13,7 +13,7 @@
// CHECK: llvm.return
// CHECK-NOT: func.func
-func.func @_QPfoo() {
+func.func @_QPfoo() -> !fir.ref<i32> {
%1 = fir.alloca i32
- return
+ return %1 : !fir.ref<i32>
}
diff --git a/flang/test/Driver/tco-test-gen.fir b/flang/test/Driver/tco-test-gen.fir
index 38d4e50ecf3aa..15483f7ee3534 100644
--- a/flang/test/Driver/tco-test-gen.fir
+++ b/flang/test/Driver/tco-test-gen.fir
@@ -42,11 +42,10 @@ func.func @_QPtest(%arg0: !fir.ref<i32> {fir.bindc_name = "num"}, %arg1: !fir.re
// CHECK-SAME: %[[ARG2:.*]]: !llvm.ptr {fir.bindc_name = "ub", llvm.nocapture},
// CHECK-SAME: %[[ARG3:.*]]: !llvm.ptr {fir.bindc_name = "step", llvm.nocapture}) {
+// CMPLX: %[[VAL_3:.*]] = llvm.mlir.constant(0 : index) : i64
+// CMPLX: %[[VAL_2:.*]] = llvm.mlir.constant(1 : index) : i64
// CMPLX: %[[VAL_0:.*]] = llvm.mlir.constant(1 : i64) : i64
// CMPLX: %[[VAL_1:.*]] = llvm.alloca %[[VAL_0]] x i32 {bindc_name = "i"} : (i64) -> !llvm.ptr
-// CMPLX: %[[VAL_2:.*]] = llvm.mlir.constant(1 : index) : i64
-// CMPLX: %[[VAL_3:.*]] = llvm.mlir.constant(0 : index) : i64
-// CMPLX: %[[VAL_4:.*]] = llvm.mlir.constant(1 : i64) : i64
// SIMPLE: %[[VAL_3:.*]] = llvm.mlir.constant(0 : index) : i64
// SIMPLE: %[[VAL_2:.*]] = llvm.mlir.constant(1 : index) : i64
diff --git a/flang/test/Fir/alloc-32.fir b/flang/test/Fir/alloc-32.fir
index a3cbf200c24fc..f57f6ce6fcf5e 100644
--- a/flang/test/Fir/alloc-32.fir
+++ b/flang/test/Fir/alloc-32.fir
@@ -19,7 +19,7 @@ func.func @allocmem_scalar_nonchar() -> !fir.heap<i32> {
// CHECK-LABEL: define ptr @allocmem_scalar_dynchar(
// CHECK-SAME: i32 %[[len:.*]])
// CHECK: %[[mul1:.*]] = sext i32 %[[len]] to i64
-// CHECK: %[[mul2:.*]] = mul i64 1, %[[mul1]]
+// CHECK: %[[mul2:.*]] = mul i64 %[[mul1]], 1
// CHECK: %[[cmp:.*]] = icmp sgt i64 %[[mul2]], 0
// CHECK: %[[sz:.*]] = select i1 %[[cmp]], i64 %[[mul2]], i64 1
// CHECK: %[[trunc:.*]] = trunc i64 %[[sz]] to i32
diff --git a/flang/test/Fir/alloc.fir b/flang/test/Fir/alloc.fir
index 8da8b828c18b9..0d3ce323d0d7c 100644
--- a/flang/test/Fir/alloc.fir
+++ b/flang/test/Fir/alloc.fir
@@ -86,7 +86,7 @@ func.func @alloca_scalar_dynchar_kind(%l : i32) -> !fir.ref<!fir.char<2,?>> {
// CHECK-LABEL: define ptr @allocmem_scalar_dynchar(
// CHECK-SAME: i32 %[[len:.*]])
// CHECK: %[[mul1:.*]] = sext i32 %[[len]] to i64
-// CHECK: %[[mul2:.*]] = mul i64 1, %[[mul1]]
+// CHECK: %[[mul2:.*]] = mul i64 %[[mul1]], 1
// CHECK: %[[cmp:.*]] = icmp sgt i64 %[[mul2]], 0
// CHECK: %[[size:.*]] = select i1 %[[cmp]], i64 %[[mul2]], i64 1
// CHECK: call ptr @malloc(i64 %[[size]])
@@ -98,7 +98,7 @@ func.func @allocmem_scalar_dynchar(%l : i32) -> !fir.heap<!fir.char<1,?>> {
// CHECK-LABEL: define ptr @allocmem_scalar_dynchar_kind(
// CHECK-SAME: i32 %[[len:.*]])
// CHECK: %[[mul1:.*]] = sext i32 %[[len]] to i64
-// CHECK: %[[mul2:.*]] = mul i64 2, %[[mul1]]
+// CHECK: %[[mul2:.*]] = mul i64 %[[mul1]], 2
// CHECK: %[[cmp:.*]] = icmp sgt i64 %[[mul2]], 0
// CHECK: %[[size:.*]] = select i1 %[[cmp]], i64 %[[mul2]], i64 1
// CHECK: call ptr @malloc(i64 %[[size]])
@@ -185,7 +185,7 @@ func.func @alloca_dynarray_of_nonchar2(%e: index) -> !fir.ref<!fir.array<?x?xi32
// CHECK-LABEL: define ptr @allocmem_dynarray_of_nonchar(
// CHECK-SAME: i64 %[[extent:.*]])
-// CHECK: %[[prod1:.*]] = mul i64 12, %[[extent]]
+// CHECK: %[[prod1:.*]] = mul i64 %[[extent]], 12
// CHECK: %[[cmp:.*]] = icmp sgt i64 %[[prod1]], 0
// CHECK: %[[size:.*]] = select i1 %[[cmp]], i64 %[[prod1]], i64 1
// CHECK: call ptr @malloc(i64 %[[size]])
@@ -196,7 +196,7 @@ func.func @allocmem_dynarray_of_nonchar(%e: index) -> !fir.heap<!fir.array<3x?xi
// CHECK-LABEL: define ptr @allocmem_dynarray_of_nonchar2(
// CHECK-SAME: i64 %[[extent:.*]])
-// CHECK: %[[prod1:.*]] = mul i64 4, %[[extent]]
+// CHECK: %[[prod1:.*]] = mul i64 %[[extent]], 4
// CHECK: %[[prod2:.*]] = mul i64 %[[prod1]], %[[extent]]
// CHECK: %[[cmp:.*]] = icmp sgt i64 %[[prod2]], 0
// CHECK: %[[size:.*]] = select i1 %[[cmp]], i64 %[[prod2]], i64 1
@@ -227,7 +227,7 @@ func.func @alloca_dynarray_of_char2(%e : index) -> !fir.ref<!fir.array<?x?x!fir.
// CHECK-LABEL: define ptr @allocmem_dynarray_of_char(
// CHECK-SAME: i64 %[[extent:.*]])
-// CHECK: %[[prod1:.*]] = mul i64 60, %[[extent]]
+// CHECK: %[[prod1:.*]] = mul i64 %[[extent]], 60
// CHECK: %[[cmp:.*]] = icmp sgt i64 %[[prod1]], 0
// CHECK: %[[size:.*]] = select i1 %[[cmp]], i64 %[[prod1]], i64 1
// CHECK: call ptr @malloc(i64 %[[size]])
@@ -238,7 +238,7 @@ func.func @allocmem_dynarray_of_char(%e : index) -> !fir.heap<!fir.array<3x?x!fi
// CHECK-LABEL: define ptr @allocmem_dynarray_of_char2(
// CHECK-SAME: i64 %[[extent:.*]])
-// CHECK: %[[prod1:.*]] = mul i64 20, %[[extent]]
+// CHECK: %[[prod1:.*]] = mul i64 %[[extent]], 20
// CHECK: %[[prod2:.*]] = mul i64 %[[prod1]], %[[extent]]
// CHECK: %[[cmp:.*]] = icmp sgt i64 %[[prod2]], 0
// CHECK: %[[size:.*]] = select i1 %[[cmp]], i64 %[[mul2]], i64 1
@@ -286,7 +286,7 @@ func.func @allocmem_dynarray_of_dynchar(%l: i32, %e : index) -> !fir.heap<!fir.a
// CHECK-LABEL: define ptr @allocmem_dynarray_of_dynchar2(
// CHECK-SAME: i32 %[[len:.*]], i64 %[[extent:.*]])
// CHECK: %[[a:.*]] = sext i32 %[[len]] to i64
-// CHECK: %[[prod1:.*]] = mul i64 2, %[[a]]
+// CHECK: %[[prod1:.*]] = mul i64 %[[a]], 2
// CHECK: %[[prod2:.*]] = mul i64 %[[prod1]], %[[extent]]
// CHECK: %[[prod3:.*]] = mul i64 %[[prod2]], %[[extent]]
// CHECK: %[[cmp:.*]] = icmp sgt i64 %[[prod3]], 0
@@ -366,12 +366,13 @@ func.func @allocmem_array_with_holes_dynchar(%arg0: index, %arg1: index) -> !fir
// CHECK: %[[VAL_0:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, ptr, [1 x i64] }, i64 1
// CHECK: %[[VAL_3:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]], ptr, [1 x i64] }, i64 1
// CHECK: %[[VAL_2:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, ptr, [1 x i64] }, i64 1
-
+func.func private @foo(%0: !fir.ref<!fir.class<none>>, %1: !fir.ref<!fir.class<!fir.array<?xnone>>>, %2: !fir.ref<!fir.box<none>>, %3: !fir.ref<!fir.box<!fir.array<?xnone>>>)
func.func @alloca_unlimited_polymorphic_box() {
%0 = fir.alloca !fir.class<none>
%1 = fir.alloca !fir.class<!fir.array<?xnone>>
%2 = fir.alloca !fir.box<none>
%3 = fir.alloca !fir.box<!fir.array<?xnone>>
+ fir.call @foo(%0, %1, %2, %3) : (!fir.ref<!fir.class<none>>, !fir.ref<!fir.class<!fir.array<?xnone>>>, !fir.ref<!fir.box<none>>, !fir.ref<!fir.box<!fir.array<?xnone>>>) -> ()
return
}
// Note: allocmem of fir.box are not possible (fir::HeapType::verify does not
diff --git a/flang/test/Fir/arrexp.fir b/flang/test/Fir/arrexp.fir
index e8ec8ac79e0c2..2eb717228d998 100644
--- a/flang/test/Fir/arrexp.fir
+++ b/flang/test/Fir/arrexp.fir
@@ -143,9 +143,9 @@ func.func @f6(%arg0: !fir.box<!fir.array<?xf32>>, %arg1: f32) {
%c9 = arith.constant 9 : index
%c10 = arith.constant 10 : index
- // CHECK: %[[EXT_GEP:.*]] = getelementptr {{.*}} %[[A]], i32 0, i32 7, i64 0, i32 1
+ // CHECK: %[[EXT_GEP:.*]] = getelementptr {{.*}} %[[A]], i32 0, i32 7, i32 0, i32 1
// CHECK: %[[EXTENT:.*]] = load i64, ptr %[[EXT_GEP]]
- // CHECK: %[[SIZE:.*]] = mul i64 4, %[[EXTENT]]
+ // CHECK: %[[SIZE:.*]] = mul i64 %[[EXTENT]], 4
// CHECK: %[[CMP:.*]] = icmp sgt i64 %[[SIZE]], 0
// CHECK: %[[SZ:.*]] = select i1 %[[CMP]], i64 %[[SIZE]], i64 1
// CHECK: %[[MALLOC:.*]] = call ptr @malloc(i64 %[[SZ]])
diff --git a/flang/test/Fir/basic-program.fir b/flang/test/Fir/basic-program.fir
index c9fe53bf093a1..6bad03dded24d 100644
--- a/flang/test/Fir/basic-program.fir
+++ b/flang/test/Fir/basic-program.fir
@@ -158,4 +158,6 @@ func.func @_QQmain() {
// PASSES-NEXT: LowerNontemporalPass
// PASSES-NEXT: FIRToLLVMLowering
// PASSES-NEXT: ReconcileUnrealizedCasts
+// PASSES-NEXT: 'llvm.func' Pipeline
+// PASSES-NEXT: PrepareForOMPOffloadPrivatizationPass
// PASSES-NEXT: LLVMIRLoweringPass
diff --git a/flang/test/Fir/box.fir b/flang/test/Fir/box.fir
index c0cf3d8375983..760fbd4792122 100644
--- a/flang/test/Fir/box.fir
+++ b/flang/test/Fir/box.fir
@@ -57,7 +57,7 @@ func.func @fa(%a : !fir.ref<!fir.array<100xf32>>) {
// CHECK-SAME: ptr {{[^%]*}}%[[res:.*]], ptr {{[^%]*}}%[[arg0:.*]], i64 %[[arg1:.*]])
func.func @b1(%arg0 : !fir.ref<!fir.char<1,?>>, %arg1 : index) -> !fir.box<!fir.char<1,?>> {
// CHECK: %[[alloca:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8 }
- // CHECK: %[[size:.*]] = mul i64 1, %[[arg1]]
+ // CHECK: %[[size:.*]] = mul i64 %[[arg1]], 1
// CHECK: insertvalue {{.*}} undef, i64 %[[size]], 1
// CHECK: insertvalue {{.*}} i32 20240719, 2
// CHECK: insertvalue {{.*}} ptr %[[arg0]], 0
@@ -89,7 +89,7 @@ func.func @b2(%arg0 : !fir.ref<!fir.array<?x!fir.char<1,5>>>, %arg1 : index) ->
func.func @b3(%arg0 : !fir.ref<!fir.array<?x!fir.char<1,?>>>, %arg1 : index, %arg2 : index) -> !fir.box<!fir.array<?x!fir.char<1,?>>> {
%1 = fir.shape %arg2 : (index) -> !fir.shape<1>
// CHECK: %[[alloca:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }
- // CHECK: %[[size:.*]] = mul i64 1, %[[arg1]]
+ // CHECK: %[[size:.*]] = mul i64 %[[arg1]], 1
// CHECK: insertvalue {{.*}} i64 %[[size]], 1
// CHECK: insertvalue {{.*}} i32 20240719, 2
// CHECK: insertvalue {{.*}} i64 %[[arg2]], 7, 0, 1
@@ -108,7 +108,7 @@ func.func @b4(%arg0 : !fir.ref<!fir.array<7x!fir.char<1,?>>>, %arg1 : index) ->
%c_7 = arith.constant 7 : index
%1 = fir.shape %c_7 : (index) -> !fir.shape<1>
// CHECK: %[[alloca:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }
- // CHECK: %[[size:.*]] = mul i64 1, %[[arg1]]
+ // CHECK: %[[size:.*]] = mul i64 %[[arg1]], 1
// CHECK: insertvalue {{.*}} i64 %[[size]], 1
// CHECK: insertvalue {{.*}} i32 20240719, 2
// CHECK: insertvalue {{.*}} i64 7, 7, 0, 1
diff --git a/flang/test/Fir/boxproc.fir b/flang/test/Fir/boxproc.fir
index 97d9b38ed6f40..d4c36a4f5b213 100644
--- a/flang/test/Fir/boxproc.fir
+++ b/flang/test/Fir/boxproc.fir
@@ -82,12 +82,8 @@ func.func @_QPtest_proc_dummy_other(%arg0: !fir.boxproc<() -> ()>) {
// CHECK: store [1 x i8] c" ", ptr %[[VAL_18]], align 1
// CHECK: call void @llvm.init.trampoline(ptr %[[VAL_20]], ptr @_QFtest_proc_dummy_charPgen_message, ptr %[[VAL_2]])
// CHECK: %[[VAL_23:.*]] = call ptr @llvm.adjust.trampoline(ptr %[[VAL_20]])
-// CHECK: %[[VAL_25:.*]] = insertvalue { ptr, i64 } undef, ptr %[[VAL_23]], 0
-// CHECK: %[[VAL_26:.*]] = insertvalue { ptr, i64 } %[[VAL_25]], i64 10, 1
// CHECK: %[[VAL_27:.*]] = call ptr @llvm.stacksave.p0()
-// CHECK: %[[VAL_28:.*]] = extractvalue { ptr, i64 } %[[VAL_26]], 0
-// CHECK: %[[VAL_29:.*]] = extractvalue { ptr, i64 } %[[VAL_26]], 1
-// CHECK: %[[VAL_30:.*]] = call { ptr, i64 } @_QPget_message(ptr %[[VAL_0]], i64 40, ptr %[[VAL_28]], i64 %[[VAL_29]])
+// CHECK: %[[VAL_30:.*]] = call { ptr, i64 } @_QPget_message(ptr %[[VAL_0]], i64 40, ptr %[[VAL_23]], i64 10)
// CHECK: %[[VAL_32:.*]] = call i1 @_FortranAioOutputAscii(ptr %{{.*}}, ptr %[[VAL_0]], i64 40)
// CHECK: call void @llvm.stackrestore.p0(ptr %[[VAL_27]])
@@ -115,14 +111,10 @@ func.func @_QPtest_proc_dummy_other(%arg0: !fir.boxproc<() -> ()>) {
// CHECK-LABEL: define { ptr, i64 } @_QPget_message(ptr
// CHECK-SAME: %[[VAL_0:.*]], i64 %[[VAL_1:.*]], ptr %[[VAL_2:.*]], i64
// CHECK-SAME: %[[VAL_3:.*]])
-// CHECK: %[[VAL_4:.*]] = insertvalue { ptr, i64 } undef, ptr %[[VAL_2]], 0
-// CHECK: %[[VAL_5:.*]] = insertvalue { ptr, i64 } %[[VAL_4]], i64 %[[VAL_3]], 1
-// CHECK: %[[VAL_7:.*]] = extractvalue { ptr, i64 } %[[VAL_5]], 0
-// CHECK: %[[VAL_8:.*]] = extractvalue { ptr, i64 } %[[VAL_5]], 1
// CHECK: %[[VAL_9:.*]] = call ptr @llvm.stacksave.p0()
-// CHECK: %[[VAL_10:.*]] = alloca i8, i64 %[[VAL_8]], align 1
-// CHECK: %[[VAL_12:.*]] = call { ptr, i64 } %[[VAL_7]](ptr %[[VAL_10]], i64 %[[VAL_8]])
-// CHECK: %[[VAL_13:.*]] = add i64 %[[VAL_8]], 12
+// CHECK: %[[VAL_10:.*]] = alloca i8, i64 %[[VAL_3]], align 1
+// CHECK: %[[VAL_12:.*]] = call { ptr, i64 } %[[VAL_2]](ptr %[[VAL_10]], i64 %[[VAL_3]])
+// CHECK: %[[VAL_13:.*]] = add i64 %[[VAL_3]], 12
// CHECK: %[[VAL_14:.*]] = alloca i8, i64 %[[VAL_13]], align 1
// CHECK: call void @llvm.memmove.p0.p0.i64(ptr %[[VAL_14]], ptr {{.*}}, i64 12, i1 false)
// CHECK: %[[VAL_18:.*]] = phi i64
diff --git a/flang/test/Fir/embox.fir b/flang/test/Fir/embox.fir
index 0f304cff2c79e..11f7457b6873c 100644
--- a/flang/test/Fir/embox.fir
+++ b/flang/test/Fir/embox.fir
@@ -11,7 +11,7 @@ func.func @_QPtest_callee(%arg0: !fir.box<!fir.array<?xi32>>) {
func.func @_QPtest_slice() {
// CHECK: %[[a1:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }, align 8
// CHECK: %[[a2:.*]] = alloca [20 x i32], i64 1, align 4
-// CHECK: %[[a3:.*]] = getelementptr [20 x i32], ptr %[[a2]], i64 0, i64 0
+// CHECK: %[[a3:.*]] = getelementptr [20 x i32], ptr %[[a2]], i32 0, i64 0
// CHECK: %[[a4:.*]] = insertvalue { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }
// CHECK: { ptr undef, i64 4, i32 20240719, i8 1, i8 9, i8 0, i8 0, [1 x [3 x i64]]
// CHECK: [i64 1, i64 5, i64 8]] }, ptr %[[a3]], 0
@@ -38,7 +38,7 @@ func.func @_QPtest_dt_callee(%arg0: !fir.box<!fir.array<?xi32>>) {
func.func @_QPtest_dt_slice() {
// CHECK: %[[a1:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }, align 8
// CHECK: %[[a3:.*]] = alloca [20 x %_QFtest_dt_sliceTt], i64 1, align 8
-// CHECK: %[[a4:.*]] = getelementptr [20 x %_QFtest_dt_sliceTt], ptr %[[a3]], i64 0, i64 0, i32 0
+// CHECK: %[[a4:.*]] = getelementptr [20 x %_QFtest_dt_sliceTt], ptr %[[a3]], i32 0, i64 0, i32 0
// CHECK: %[[a5:.*]] = insertvalue { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }
// CHECK-SAME: { ptr undef, i64 4, i32 20240719, i8 1, i8 9, i8 0, i8 0, [1 x [3 x i64]]
// CHECK-SAME: [i64 1, i64 5, i64 16
@@ -73,7 +73,7 @@ func.func @emboxSubstring(%arg0: !fir.ref<!fir.array<2x3x!fir.char<1,4>>>) {
%0 = fir.shape %c2, %c3 : (index, index) -> !fir.shape<2>
%1 = fir.slice %c1, %c2, %c1, %c1, %c3, %c1 substr %c1_i64, %c2_i64 : (index, index, index, index, index, index, i64, i64) -> !fir.slice<2>
%2 = fir.embox %arg0(%0) [%1] : (!fir.ref<!fir.array<2x3x!fir.char<1,4>>>, !fir.shape<2>, !fir.slice<2>) -> !fir.box<!fir.array<?x?x!fir.char<1,?>>>
- // CHECK: %[[addr:.*]] = getelementptr [3 x [2 x [4 x i8]]], ptr %[[arg0]], i64 0, i64 0, i64 0, i64 1
+ // CHECK: %[[addr:.*]] = getelementptr [3 x [2 x [4 x i8]]], ptr %[[arg0]], i32 0, i64 0, i64 0, i32 1
// CHECK: insertvalue {[[descriptorType:.*]]} { ptr undef, i64 2, i32 20240719, i8 2, i8 40, i8 0, i8 0
// CHECK-SAME: [2 x [3 x i64]] [{{\[}}3 x i64] [i64 1, i64 2, i64 4], [3 x i64] [i64 1, i64 3, i64 8]] }
// CHECK-SAME: ptr %[[addr]], 0
diff --git a/flang/test/Fir/omp-reduction-embox-codegen.fir b/flang/test/Fir/omp-reduction-embox-codegen.fir
index 1645e1a407ad4..e517b1352ff5c 100644
--- a/flang/test/Fir/omp-reduction-embox-codegen.fir
+++ b/flang/test/Fir/omp-reduction-embox-codegen.fir
@@ -23,14 +23,14 @@ omp.declare_reduction @test_reduction : !fir.ref<!fir.box<i32>> init {
omp.yield(%0 : !fir.ref<!fir.box<i32>>)
}
-func.func @_QQmain() attributes {fir.bindc_name = "reduce"} {
+func.func @_QQmain() -> !fir.ref<!fir.box<i32>> attributes {fir.bindc_name = "reduce"} {
%4 = fir.alloca !fir.box<i32>
omp.parallel reduction(byref @test_reduction %4 -> %arg0 : !fir.ref<!fir.box<i32>>) {
omp.terminator
}
- return
+ return %4: !fir.ref<!fir.box<i32>>
}
// basically we are testing that there isn't a crash
-// CHECK-LABEL: define void @_QQmain
+// CHECK-LABEL: define ptr @_QQmain
// CHECK-NEXT: alloca { ptr, i64, i32, i8, i8, i8, i8 }, i64 1, align 8
diff --git a/flang/test/Fir/optional.fir b/flang/test/Fir/optional.fir
index bded8b5332a30..66ff69f083467 100644
--- a/flang/test/Fir/optional.fir
+++ b/flang/test/Fir/optional.fir
@@ -37,8 +37,7 @@ func.func @bar2() -> i1 {
// CHECK-LABEL: @foo3
func.func @foo3(%arg0: !fir.boxchar<1>) -> i1 {
- // CHECK: %[[extract:.*]] = extractvalue { ptr, i64 } %{{.*}}, 0
- // CHECK: %[[ptr:.*]] = ptrtoint ptr %[[extract]] to i64
+ // CHECK: %[[ptr:.*]] = ptrtoint ptr %0 to i64
// CHECK: icmp ne i64 %[[ptr]], 0
%0 = fir.is_present %arg0 : (!fir.boxchar<1>) -> i1
return %0 : i1
diff --git a/flang/test/Fir/pdt.fir b/flang/test/Fir/pdt.fir
index a200cd7e7cc03..411927aae6bdf 100644
--- a/flang/test/Fir/pdt.fir
+++ b/flang/test/Fir/pdt.fir
@@ -96,13 +96,13 @@ func.func @_QTt1P.f2.offset(%0 : i32, %1 : i32) -> i32 {
func.func private @bar(!fir.ref<!fir.char<1,?>>)
-// CHECK-LABEL: define void @_QPfoo(i32 %0, i32 %1)
-func.func @_QPfoo(%arg0 : i32, %arg1 : i32) {
+// CHECK-LABEL: define ptr @_QPfoo(i32 %0, i32 %1)
+func.func @_QPfoo(%arg0 : i32, %arg1 : i32) -> !fir.ref<!fir.type<_QTt1>> {
// CHECK: %[[size:.*]] = call i64 @_QTt1P.mem.size(i32 %0, i32 %1)
// CHECK: %[[alloc:.*]] = alloca i8, i64 %[[size]]
%0 = fir.alloca !fir.type<_QTt1(p1:i32,p2:i32){f1:!fir.char<1,?>,f2:!fir.char<1,?>}>(%arg0, %arg1 : i32, i32)
//%2 = fir.coordinate_of %0, f2 : (!fir.ref<!fir.type<_QTt1>>) -> !fir.ref<!fir.char<1,?>>
%2 = fir.zero_bits !fir.ref<!fir.char<1,?>>
fir.call @bar(%2) : (!fir.ref<!fir.char<1,?>>) -> ()
- return
+ return %0 : !fir.ref<!fir.type<_QTt1>>
}
diff --git a/flang/test/Fir/rebox.fir b/flang/test/Fir/rebox.fir
index 0c9f6d9bb94ad..d858adfb7c45d 100644
--- a/flang/test/Fir/rebox.fir
+++ b/flang/test/Fir/rebox.fir
@@ -36,7 +36,7 @@ func.func @test_rebox_1(%arg0: !fir.box<!fir.array<?x?xf32>>) {
// CHECK: %[[VOIDBASE0:.*]] = getelementptr i8, ptr %[[INBASE]], i64 %[[OFFSET_0]]
// CHECK: %[[OFFSET_1:.*]] = mul i64 2, %[[INSTRIDE_1]]
// CHECK: %[[VOIDBASE1:.*]] = getelementptr i8, ptr %[[VOIDBASE0]], i64 %[[OFFSET_1]]
- // CHECK: %[[OUTSTRIDE0:.*]] = mul i64 3, %[[INSTRIDE_1]]
+ // CHECK: %[[OUTSTRIDE0:.*]] = mul i64 %[[INSTRIDE_1]], 3
// CHECK: %[[OUTBOX1:.*]] = insertvalue { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] } %{{.*}}, i64 %[[OUTSTRIDE0]], 7, 0, 2
// CHECK: %[[OUTBOX2:.*]] = insertvalue { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] } %[[OUTBOX1]], ptr %[[VOIDBASE1]], 0
// CHECK: store { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] } %[[OUTBOX2]], ptr %[[OUTBOX_ALLOC]], align 8
@@ -63,7 +63,7 @@ func.func @test_rebox_2(%arg0: !fir.box<!fir.array<?x?x!fir.char<1,?>>>) {
// CHECK: %[[OUTBOX:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, [2 x [3 x i64]] }
// CHECK: %[[LEN_GEP:.*]] = getelementptr { ptr, i64, i32, i8, i8, i8, i8, [2 x [3 x i64]] }, ptr %[[INBOX]], i32 0, i32 1
// CHECK: %[[LEN:...
[truncated]
|
|
✅ With the latest revision this PR passed the C/C++ code formatter. |
tblah
reviewed
Aug 26, 2025
Contributor
tblah
left a comment
tblah
left a comment
There was a problem hiding this comment.
Why did you decide to do this in a pass instead of handling it in MLIR -> LLVM conversion as is done for omp task?
Contributor
Author
I did anticipate this question especially because MLIR -> LLVM translation is where I had first started out with the intention of extending your work on A couple of reasons make it too late to do this during MLIR - LLVMIR translation. Too late as in not impossible, but arguably harder to get correct and maintain thereafter. Essentially, what we need to do is
Now, to allocate heap memory for the private variable, we'd have two options
This requires addl bookkeeping and coordination between |
…get-tasks
This patch adds support for translation of the private clause on deferred
target tasks - that is `omp.target` operations with the `nowait` clause.
An offloading call for a deferred target-task is not blocking - the offloading
host task continues it execution after issuing the offloading call. Therefore,
the key problem we need to solve is to ensure that the data needed for private
variables to be initialized in the target task persists even after the host
task has completed.
We do this in a new pass called PrepareForOMPOffloadPrivatizationPass. For a privatized
variable that needs its host counterpart for initialization (such as the shape
of the data from the descriptor when an allocatable is privatized or the value of
the data when an allocatable is firstprivatized),
- the pass allocates memory on the heap.
- it then initializes this memory by copying the contents of host variable to
the newly allocated location on the heap.
- Then, the pass updates all the `omp.map.info` operations that pointed to the
host variable to now point to the one located in the heap.
The pass uses a rewrite pattern applied using the greedy pattern matcher, which
in turn does some constant folding and DCE. Due to this a number of lit tests
had to be updated. In GEPs constant get folded into indices and truncated to
i32 types. In some tests sequence of insertvalue and extractvalue instructions
get cancelled out. So, these needed to be updated too.
…reedy pattern matcher
bhandarkar-pranav
force-pushed
the
flang/delayed_priv_def_tgt_tasks_translation
branch
from
ff8afbd to
c859bbc
Compare
Contributor
|
Ahh I see what you mean. This is different because as well as being (first)private, these variables may also be mapped, which adds another layer of complexity. I haven't followed much about mapping so I will take your word for it and leave it for experts in offloading to give their opinions. |
tblah
reviewed
Aug 27, 2025
mlir/lib/Dialect/LLVMIR/Transforms/OpenMPOffloadPrivatizationPrepare.cpp
Outdated
Show resolved
Hide resolved
mlir/lib/Dialect/LLVMIR/Transforms/OpenMPOffloadPrivatizationPrepare.cpp
Outdated
Show resolved
Hide resolved
|
|
||
| // Allocate heap memory that corresponds to the type of memory | ||
| // pointed to by varPtr | ||
| // TODO: For boxchars this likely wont be a pointer. |
Contributor
There was a problem hiding this comment.
Yes you can see in the code for tasks that boxchars are a hack because you can't really have a !fir.ref<!fir.boxchar<>> in the FIR type system. This is handled for tasks so you can see what I did there.
Contributor
Author
There was a problem hiding this comment.
Thank you, I'll take a look.
| // Copy the value of the local variable into the heap-allocated location. | ||
| mlir::Location loc = chainOfOps.front()->getLoc(); | ||
| mlir::Type varType = getElemType(varPtr); | ||
| auto loadVal = rewriter.create<LLVM::LoadOp>(loc, varType, varPtr); |
Contributor
There was a problem hiding this comment.
What about more complex types e.g. arrays, derived types?
For firstprivate you can use the copy region in the privatizer. For plain private you just need to use an init region to initialise non-trivial types but don't need to copy. This initialisation and copying must happen synchronously.
Contributor
Author
There was a problem hiding this comment.
You are right that there is a problem here. The problem is that derived types could have any number of pointers and therefore deep copies will be needed. Are you suggesting that I use the copy region of the privatizer by cloning it to get the deep copy that i need? Before that though, I'd have to allocate memory for each pointer inside the derived type. I was hoping to tackle derived types in a subsequent PR, which I should have made clear in this PR.
Contributor
There was a problem hiding this comment.
If you intend to handle these in a later PR please catch this and create a useful error message (similar to the not yet implemented messages generated from OpenMP MLIR dialect to LLVM IR conversion.
Yes you can do initial allocation/initialization for the derived type by inlining the init region from the privatizer. If it is first private then you can inline the copy region to get a type appropriate copy. So far as I know init+copy should be sufficient without any extra allocation.
mlir/lib/Dialect/LLVMIR/Transforms/OpenMPOffloadPrivatizationPrepare.cpp
Outdated
Show resolved
Hide resolved
mlir/test/Dialect/LLVMIR/omp-offload-privatization-prepare.mlir
Outdated
Show resolved
Hide resolved
Meinersbur
reviewed
Aug 28, 2025
mlir/lib/Dialect/LLVMIR/Transforms/OpenMPOffloadPrivatizationPrepare.cpp
Outdated
Show resolved
Hide resolved
mlir/lib/Dialect/LLVMIR/Transforms/OpenMPOffloadPrivatizationPrepare.cpp
Outdated
Show resolved
Hide resolved
mlir/lib/Dialect/LLVMIR/Transforms/OpenMPOffloadPrivatizationPrepare.cpp
Outdated
Show resolved
Hide resolved
mlir/include/mlir/Dialect/LLVMIR/Transforms/OpenMPOffloadPrivatizationPrepare.h
Outdated
Show resolved
Hide resolved
mlir/include/mlir/Dialect/LLVMIR/Transforms/OpenMPOffloadPrivatizationPrepare.h
Outdated
Show resolved
Hide resolved
mlir/lib/Dialect/LLVMIR/Transforms/OpenMPOffloadPrivatizationPrepare.cpp
Outdated
Show resolved
Hide resolved
mlir/lib/Dialect/LLVMIR/Transforms/OpenMPOffloadPrivatizationPrepare.cpp
Outdated
Show resolved
Hide resolved
mlir/lib/Dialect/LLVMIR/Transforms/OpenMPOffloadPrivatizationPrepare.cpp
Outdated
Show resolved
Hide resolved
ergawy
reviewed
Oct 16, 2025
Member
ergawy
left a comment
ergawy
left a comment
There was a problem hiding this comment.
Sorry for the delay here Pranav. Went through the whole PR. Awesome work, just a number of not so large comments.
mlir/include/mlir/Dialect/OpenMP/Transforms/OpenMPOffloadPrivatizationPrepare.h
Outdated
Show resolved
Hide resolved
mlir/lib/Dialect/OpenMP/Transforms/OpenMPOffloadPrivatizationPrepare.cpp
Outdated
Show resolved
Hide resolved
mlir/lib/Dialect/OpenMP/Transforms/OpenMPOffloadPrivatizationPrepare.cpp
Outdated
Show resolved
Hide resolved
mlir/lib/Dialect/OpenMP/Transforms/OpenMPOffloadPrivatizationPrepare.cpp
Outdated
Show resolved
Hide resolved
mlir/lib/Dialect/OpenMP/Transforms/OpenMPOffloadPrivatizationPrepare.cpp
Outdated
Show resolved
Hide resolved
Contributor
Author
No problem at all, thank you for the review. I'll take care of your comments today. |
…erating on FuncOp to one operating on ModuleOp.
Contributor
Author
ergawy
approved these changes
Oct 22, 2025
Member
ergawy
left a comment
ergawy
left a comment
There was a problem hiding this comment.
LGTM! Thanks for taking care of my comments Pranav.
tblah
approved these changes
Oct 22, 2025
Collaborator
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/169/builds/16264 Here is the relevant piece of the build log for the reference |
thurstond
reviewed
Oct 22, 2025
| rewriter.setInsertionPoint(cloneModifyAndErase(mapInfoOp)); | ||
|
|
||
| // Fix any members that may use varPtr to now use heapMem | ||
| for (auto member : mapInfoOp.getMembers()) { |
Contributor
There was a problem hiding this comment.
I think there is a heap use-after-free here: mapInfoOp should not be accessed here because cloneModifyAndErase (line 255) erased it.
(This was identified by a buildbot: https://lab.llvm.org/buildbot/#/builders/52/builds/12179/steps/12/logs/stdio)
cc @fmayer
thurstond
added a commit
to thurstond/llvm-project
that referenced
this pull request
Oct 22, 2025
mapInfoOp should not be accessed here because cloneModifyAndErase (line 255) erased it. Fix the issue by replacing mapInfoOp with the cloned op.
thurstond
added a commit
that referenced
this pull request
Oct 22, 2025
`mapInfoOp.getMembers()` on line 258 is use-after-free, because cloneModifyAndErase (line 255) erased `mapInfoOp`. Fix the issue by replacing subsequent `mapInfoOp` usages with `clonedOp`. Similarly, update `memberMapInfoOp` to avoid subsequent use-after-free.
mikolaj-pirog
pushed a commit
to mikolaj-pirog/llvm-project
that referenced
this pull request
Oct 23, 2025
…4712) `mapInfoOp.getMembers()` on line 258 is use-after-free, because cloneModifyAndErase (line 255) erased `mapInfoOp`. Fix the issue by replacing subsequent `mapInfoOp` usages with `clonedOp`. Similarly, update `memberMapInfoOp` to avoid subsequent use-after-free.
cota
added a commit
to cota/llvm-project
that referenced
this pull request
Oct 23, 2025
…llvm#155348 This fixes strict weak ordering checks violations from llvm#155348 when running these two tests: mlir/test/Dialect/OpenMP/omp-offload-privatization-prepare.mlir mlir/test/Dialect/OpenMP/omp-offload-privatization-prepare-by-value.mlir Sample error: /stable/src/libcxx/include/__debug_utils/strict_weak_ordering_check.h:50: libc++ Hardening assertion !__comp(*(__first + __a), *(__first + __b)) failed: Your comparator is not a valid strict-weak ordering This is because (x < x) should be false, not true, to meet the irreflexibility property. (Note that .dominates(x, x) returns true.) I'm afraid that even after this commit we can't guarantee a strict weak ordering, because we can't guarantee transitivity of equivalence by sorting with a strict dominance function. However the tests are not failing anymore, and I am not at all familiar with this code so I will leave this concern up to the original author for consideration. (Ideas without any further context: I would consider a topological sort or walking a dominator tree.) Reference on std::sort and strict weak ordering: https://danlark.org/2022/04/20/changing-stdsort-at-googles-scale-and-beyond/
cota
added a commit
that referenced
this pull request
Oct 24, 2025
…#155348 (#164833) This fixes strict weak ordering checks violations from #155348 when running these two tests: mlir/test/Dialect/OpenMP/omp-offload-privatization-prepare.mlir mlir/test/Dialect/OpenMP/omp-offload-privatization-prepare-by-value.mlir Sample error: /stable/src/libcxx/include/__debug_utils/strict_weak_ordering_check.h:50: libc++ Hardening assertion !__comp(*__first + __a), *(__first + __b)) failed: Your comparator is not a valid strict-weak ordering This is because (x < x) should be false, not true, to meet the irreflexibility property. (Note that .dominates(x, x) returns true.) I'm afraid that even after this commit we can't guarantee a strict weak ordering, because we can't guarantee transitivity of equivalence by sorting with a strict dominance function. However the tests are not failing anymore, and I am not at all familiar with this code so I will leave this concern up to the original author for consideration. (Ideas without any further context: I would consider a topological sort or walking a dominator tree.) Reference on std::sort and strict weak ordering: https://danlark.org/2022/04/20/changing-stdsort-at-googles-scale-and-beyond/
dvbuka
pushed a commit
to dvbuka/llvm-project
that referenced
this pull request
Oct 27, 2025
…get-tasks (llvm#155348) This PR adds support for translation of the private clause on deferred target tasks - that is `omp.target` operations with the `nowait` clause. An offloading call for a deferred target-task is not blocking - the offloading (target-generating) host task continues its execution after issuing the offloading call. Therefore, the key problem we need to solve is to ensure that the data needed for private variables to be initialized in the target task persists even after the host task has completed. We do this in a new pass called `PrepareForOMPOffloadPrivatizationPass`. For a privatized variable that needs its host counterpart for initialization (such as the shape of the data from the descriptor when an allocatable is privatized or the value of the data when an allocatable is firstprivatized), - the pass allocates memory on the heap. - it then initializes this memory by using the `init` and `copy` (for firstprivate) regions of the corresponding `omp::PrivateClauseOp`. - Finally the memory allocated on the heap is freed using the `dealloc` region of the same `omp::PrivateClauseOp` instance. This step is not straightforward though, because we cannot simply free the memory that's going to be used by another thread without any synchronization. So, for deallocation, we create a `omp.task` after the `omp.target` and synchronize the two with a dummy dependency (using the `depend` clause). In this newly created `omp.task` we do the deallocation.
dvbuka
pushed a commit
to dvbuka/llvm-project
that referenced
this pull request
Oct 27, 2025
…4712) `mapInfoOp.getMembers()` on line 258 is use-after-free, because cloneModifyAndErase (line 255) erased `mapInfoOp`. Fix the issue by replacing subsequent `mapInfoOp` usages with `clonedOp`. Similarly, update `memberMapInfoOp` to avoid subsequent use-after-free.
dvbuka
pushed a commit
to dvbuka/llvm-project
that referenced
this pull request
Oct 27, 2025
…llvm#155348 (llvm#164833) This fixes strict weak ordering checks violations from llvm#155348 when running these two tests: mlir/test/Dialect/OpenMP/omp-offload-privatization-prepare.mlir mlir/test/Dialect/OpenMP/omp-offload-privatization-prepare-by-value.mlir Sample error: /stable/src/libcxx/include/__debug_utils/strict_weak_ordering_check.h:50: libc++ Hardening assertion !__comp(*__first + __a), *(__first + __b)) failed: Your comparator is not a valid strict-weak ordering This is because (x < x) should be false, not true, to meet the irreflexibility property. (Note that .dominates(x, x) returns true.) I'm afraid that even after this commit we can't guarantee a strict weak ordering, because we can't guarantee transitivity of equivalence by sorting with a strict dominance function. However the tests are not failing anymore, and I am not at all familiar with this code so I will leave this concern up to the original author for consideration. (Ideas without any further context: I would consider a topological sort or walking a dominator tree.) Reference on std::sort and strict weak ordering: https://danlark.org/2022/04/20/changing-stdsort-at-googles-scale-and-beyond/
Lukacma
pushed a commit
to Lukacma/llvm-project
that referenced
this pull request
Oct 29, 2025
…get-tasks (llvm#155348) This PR adds support for translation of the private clause on deferred target tasks - that is `omp.target` operations with the `nowait` clause. An offloading call for a deferred target-task is not blocking - the offloading (target-generating) host task continues its execution after issuing the offloading call. Therefore, the key problem we need to solve is to ensure that the data needed for private variables to be initialized in the target task persists even after the host task has completed. We do this in a new pass called `PrepareForOMPOffloadPrivatizationPass`. For a privatized variable that needs its host counterpart for initialization (such as the shape of the data from the descriptor when an allocatable is privatized or the value of the data when an allocatable is firstprivatized), - the pass allocates memory on the heap. - it then initializes this memory by using the `init` and `copy` (for firstprivate) regions of the corresponding `omp::PrivateClauseOp`. - Finally the memory allocated on the heap is freed using the `dealloc` region of the same `omp::PrivateClauseOp` instance. This step is not straightforward though, because we cannot simply free the memory that's going to be used by another thread without any synchronization. So, for deallocation, we create a `omp.task` after the `omp.target` and synchronize the two with a dummy dependency (using the `depend` clause). In this newly created `omp.task` we do the deallocation.
Lukacma
pushed a commit
to Lukacma/llvm-project
that referenced
this pull request
Oct 29, 2025
…4712) `mapInfoOp.getMembers()` on line 258 is use-after-free, because cloneModifyAndErase (line 255) erased `mapInfoOp`. Fix the issue by replacing subsequent `mapInfoOp` usages with `clonedOp`. Similarly, update `memberMapInfoOp` to avoid subsequent use-after-free.
Lukacma
pushed a commit
to Lukacma/llvm-project
that referenced
this pull request
Oct 29, 2025
…llvm#155348 (llvm#164833) This fixes strict weak ordering checks violations from llvm#155348 when running these two tests: mlir/test/Dialect/OpenMP/omp-offload-privatization-prepare.mlir mlir/test/Dialect/OpenMP/omp-offload-privatization-prepare-by-value.mlir Sample error: /stable/src/libcxx/include/__debug_utils/strict_weak_ordering_check.h:50: libc++ Hardening assertion !__comp(*__first + __a), *(__first + __b)) failed: Your comparator is not a valid strict-weak ordering This is because (x < x) should be false, not true, to meet the irreflexibility property. (Note that .dominates(x, x) returns true.) I'm afraid that even after this commit we can't guarantee a strict weak ordering, because we can't guarantee transitivity of equivalence by sorting with a strict dominance function. However the tests are not failing anymore, and I am not at all familiar with this code so I will leave this concern up to the original author for consideration. (Ideas without any further context: I would consider a topological sort or walking a dominator tree.) Reference on std::sort and strict weak ordering: https://danlark.org/2022/04/20/changing-stdsort-at-googles-scale-and-beyond/
aokblast
pushed a commit
to aokblast/llvm-project
that referenced
this pull request
Oct 30, 2025
…get-tasks (llvm#155348) This PR adds support for translation of the private clause on deferred target tasks - that is `omp.target` operations with the `nowait` clause. An offloading call for a deferred target-task is not blocking - the offloading (target-generating) host task continues its execution after issuing the offloading call. Therefore, the key problem we need to solve is to ensure that the data needed for private variables to be initialized in the target task persists even after the host task has completed. We do this in a new pass called `PrepareForOMPOffloadPrivatizationPass`. For a privatized variable that needs its host counterpart for initialization (such as the shape of the data from the descriptor when an allocatable is privatized or the value of the data when an allocatable is firstprivatized), - the pass allocates memory on the heap. - it then initializes this memory by using the `init` and `copy` (for firstprivate) regions of the corresponding `omp::PrivateClauseOp`. - Finally the memory allocated on the heap is freed using the `dealloc` region of the same `omp::PrivateClauseOp` instance. This step is not straightforward though, because we cannot simply free the memory that's going to be used by another thread without any synchronization. So, for deallocation, we create a `omp.task` after the `omp.target` and synchronize the two with a dummy dependency (using the `depend` clause). In this newly created `omp.task` we do the deallocation.
aokblast
pushed a commit
to aokblast/llvm-project
that referenced
this pull request
Oct 30, 2025
…4712) `mapInfoOp.getMembers()` on line 258 is use-after-free, because cloneModifyAndErase (line 255) erased `mapInfoOp`. Fix the issue by replacing subsequent `mapInfoOp` usages with `clonedOp`. Similarly, update `memberMapInfoOp` to avoid subsequent use-after-free.
aokblast
pushed a commit
to aokblast/llvm-project
that referenced
this pull request
Oct 30, 2025
…llvm#155348 (llvm#164833) This fixes strict weak ordering checks violations from llvm#155348 when running these two tests: mlir/test/Dialect/OpenMP/omp-offload-privatization-prepare.mlir mlir/test/Dialect/OpenMP/omp-offload-privatization-prepare-by-value.mlir Sample error: /stable/src/libcxx/include/__debug_utils/strict_weak_ordering_check.h:50: libc++ Hardening assertion !__comp(*__first + __a), *(__first + __b)) failed: Your comparator is not a valid strict-weak ordering This is because (x < x) should be false, not true, to meet the irreflexibility property. (Note that .dominates(x, x) returns true.) I'm afraid that even after this commit we can't guarantee a strict weak ordering, because we can't guarantee transitivity of equivalence by sorting with a strict dominance function. However the tests are not failing anymore, and I am not at all familiar with this code so I will leave this concern up to the original author for consideration. (Ideas without any further context: I would consider a topological sort or walking a dominator tree.) Reference on std::sort and strict weak ordering: https://danlark.org/2022/04/20/changing-stdsort-at-googles-scale-and-beyond/
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
8 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR adds support for translation of the private clause on deferred target tasks - that is
omp.targetoperations with thenowaitclause.An offloading call for a deferred target-task is not blocking - the offloading host task continues it execution after issuing the offloading call. Therefore, the key problem we need to solve is to ensure that the data needed for private variables to be initialized in the target task persists even after the host task has completed.
We do this in a new pass called PrepareForOMPOffloadPrivatizationPass. For a privatized variable that needs its host counterpart for initialization (such as the shape of the data from the descriptor when an allocatable is privatized or the value of the data when an allocatable is firstprivatized),
initandcopy(for firstprivate) regions of the correspondingomp::PrivateClauseOp.deallocregion of the sameomp::PrivateClauseOpinstance. This step is not straightforward though, because we cannot simply free the memory that's going to be used by another thread without any synchronization. So, for deallocation, we create aomp.taskafter theomp.targetand synchronize the two with a dummy dependency (using thedependclause). In this newly createdomp.taskwe do the deallocation.