intel/llvm
1
0
Fork 0
mirror of https://github.com/intel/llvm.git synced 2026-01-18 07:57:36 +08:00
Code Issues Packages Projects Releases Wiki Activity

[flang][hlfir] Codegen for unordered elemental operations.

Browse Source 
Depends on D154031, D154032

Reviewed By: jeanPerier, tblah

Differential Revision: https://reviews.llvm.org/D154033
This commit is contained in:
Slava Zakharin
2023-06-29 09:41:44 -07:00
parent 583168ee86
commit 39e87db192
4 changed files with 79 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
flang/lib/Optimizer/HLFIR/Transforms/BufferizeHLFIR.cpp
View File

@@ -557,7 +557,8 @@ struct ElementalOpConversion
adaptor.getTypeparams());
// Generate a loop nest looping around the fir.elemental shape and clone
// fir.elemental region inside the inner loop.
hlfir::LoopNest loopNest = hlfir::genLoopNest(loc, builder, extents);
hlfir::LoopNest loopNest =
hlfir::genLoopNest(loc, builder, extents, !elemental.isOrdered());
auto insPt = builder.saveInsertionPoint();
builder.setInsertionPointToStart(loopNest.innerLoop.getBody());
auto yield = hlfir::inlineElementalOp(loc, builder, elemental,

4
flang/lib/Optimizer/HLFIR/Transforms/LowerHLFIROrderedAssignments.cpp
View File

@@ -701,8 +701,8 @@ OrderedAssignmentRewriter::generateYieldedLHS(mlir::Location loc,
for (auto &op : lhsRegion.front().without_terminator())
(void)builder.clone(op, mapper);
mlir::Value newShape = mapper.lookupOrDefault(elementalAddrLhs.getShape());
loweredLhs.vectorSubscriptLoopNest =
hlfir::genLoopNest(loc, builder, newShape);
loweredLhs.vectorSubscriptLoopNest = hlfir::genLoopNest(
loc, builder, newShape, !elementalAddrLhs.isOrdered());
builder.setInsertionPointToStart(
loweredLhs.vectorSubscriptLoopNest->innerLoop.getBody());
mapper.map(elementalAddrLhs.getIndices(),

32
flang/test/HLFIR/elemental-codegen.fir
View File

@@ -107,3 +107,35 @@ func.func @derived_transpose(%arg0: !fir.box<!fir.array<?x?x!fir.type<t{field:f3
// CHECK: %[[VAL_12:.*]] = fir.undefined tuple<!fir.box<!fir.array<?x?x!fir.type<t{field:f32}>>>, i1>
// CHECK: %[[VAL_13:.*]] = fir.insert_value %[[VAL_12]], %[[VAL_6]], [1 : index] : (tuple<!fir.box<!fir.array<?x?x!fir.type<t{field:f32}>>>, i1>, i1) -> tuple<!fir.box<!fir.array<?x?x!fir.type<t{field:f32}>>>, i1>
// CHECK: %[[VAL_14:.*]] = fir.insert_value %[[VAL_13]], %[[VAL_5]]#0, [0 : index] : (tuple<!fir.box<!fir.array<?x?x!fir.type<t{field:f32}>>>, i1>, !fir.box<!fir.array<?x?x!fir.type<t{field:f32}>>>) -> tuple<!fir.box<!fir.array<?x?x!fir.type<t{field:f32}>>>, i1>
func.func @unordered() {
%c10 = arith.constant 10 : index
%c20 = arith.constant 20 : index
%0 = fir.shape %c10, %c20 : (index, index) -> !fir.shape<2>
%3 = hlfir.elemental %0 unordered : (!fir.shape<2>) -> !hlfir.expr<10x20xi32> {
^bb0(%i: index, %j: index):
%c0 = arith.constant 0 : i32
hlfir.yield_element %c0 : i32
}
return
}
// CHECK-LABEL: func.func @unordered() {
// CHECK: %[[VAL_0:.*]] = arith.constant 10 : index
// CHECK: %[[VAL_1:.*]] = arith.constant 20 : index
// CHECK: %[[VAL_2:.*]] = fir.shape %[[VAL_0]], %[[VAL_1]] : (index, index) -> !fir.shape<2>
// CHECK: %[[VAL_3:.*]] = fir.allocmem !fir.array<10x20xi32> {bindc_name = ".tmp.array", uniq_name = ""}
// CHECK: %[[VAL_4:.*]]:2 = hlfir.declare %[[VAL_3]](%[[VAL_2]]) {uniq_name = ".tmp.array"} : (!fir.heap<!fir.array<10x20xi32>>, !fir.shape<2>) -> (!fir.heap<!fir.array<10x20xi32>>, !fir.heap<!fir.array<10x20xi32>>)
// CHECK: %[[VAL_5:.*]] = arith.constant true
// CHECK: %[[VAL_6:.*]] = arith.constant 1 : index
// CHECK: fir.do_loop %[[VAL_7:.*]] = %[[VAL_6]] to %[[VAL_1]] step %[[VAL_6]] unordered {
// CHECK: fir.do_loop %[[VAL_8:.*]] = %[[VAL_6]] to %[[VAL_0]] step %[[VAL_6]] unordered {
// CHECK: %[[VAL_9:.*]] = arith.constant 0 : i32
// CHECK: %[[VAL_10:.*]] = hlfir.designate %[[VAL_4]]#0 (%[[VAL_8]], %[[VAL_7]]) : (!fir.heap<!fir.array<10x20xi32>>, index, index) -> !fir.ref<i32>
// CHECK: hlfir.assign %[[VAL_9]] to %[[VAL_10]] temporary_lhs : i32, !fir.ref<i32>
// CHECK: }
// CHECK: }
// CHECK: %[[VAL_11:.*]] = fir.undefined tuple<!fir.heap<!fir.array<10x20xi32>>, i1>
// CHECK: %[[VAL_12:.*]] = fir.insert_value %[[VAL_11]], %[[VAL_5]], [1 : index] : (tuple<!fir.heap<!fir.array<10x20xi32>>, i1>, i1) -> tuple<!fir.heap<!fir.array<10x20xi32>>, i1>
// CHECK: %[[VAL_13:.*]] = fir.insert_value %[[VAL_12]], %[[VAL_4]]#0, [0 : index] : (tuple<!fir.heap<!fir.array<10x20xi32>>, i1>, !fir.heap<!fir.array<10x20xi32>>) -> tuple<!fir.heap<!fir.array<10x20xi32>>, i1>
// CHECK: return
// CHECK: }

43
flang/test/HLFIR/order_assignments/vector-subscripts-codegen.fir
View File

@@ -169,3 +169,46 @@ func.func @where_vector_subscripts(%arg0: !fir.ref<!fir.array<10x!fir.logical<4>
// CHECK: hlfir.assign %[[VAL_3]] to %[[VAL_18]] : f32, !fir.ref<f32>
// CHECK: }
// CHECK: }
func.func @unordered(%arg0: !fir.ref<!fir.array<100xf32>> , %arg1: !fir.ref<!fir.array<10xi64>> , %arg2: !fir.ref<!fir.array<10xf32>> ) {
%c10 = arith.constant 10 : index
%c100 = arith.constant 100 : index
%0 = fir.shape %c100 : (index) -> !fir.shape<1>
%1:2 = hlfir.declare %arg0(%0) {uniq_name = "_QFsimpleEx"} : (!fir.ref<!fir.array<100xf32>>, !fir.shape<1>) -> (!fir.ref<!fir.array<100xf32>>, !fir.ref<!fir.array<100xf32>>)
%2 = fir.shape %c10 : (index) -> !fir.shape<1>
%3:2 = hlfir.declare %arg1(%2) {uniq_name = "y"} : (!fir.ref<!fir.array<10xi64>>, !fir.shape<1>) -> (!fir.ref<!fir.array<10xi64>>, !fir.ref<!fir.array<10xi64>>)
%4:2 = hlfir.declare %arg2(%2) {uniq_name = "z"} : (!fir.ref<!fir.array<10xf32>>, !fir.shape<1>) -> (!fir.ref<!fir.array<10xf32>>, !fir.ref<!fir.array<10xf32>>)
hlfir.region_assign {
hlfir.yield %4#0 : !fir.ref<!fir.array<10xf32>>
} to {
hlfir.elemental_addr %2 unordered : !fir.shape<1> {
^bb0(%arg3: index):
%5 = hlfir.designate %3#0 (%arg3) : (!fir.ref<!fir.array<10xi64>>, index) -> !fir.ref<i64>
%6 = fir.load %5 : !fir.ref<i64>
%7 = hlfir.designate %1#0 (%6) : (!fir.ref<!fir.array<100xf32>>, i64) -> !fir.ref<f32>
hlfir.yield %7 : !fir.ref<f32>
}
}
return
}
// CHECK-LABEL: func.func @unordered(
// CHECK-SAME: %[[VAL_0:.*]]: !fir.ref<!fir.array<100xf32>>,
// CHECK-SAME: %[[VAL_1:.*]]: !fir.ref<!fir.array<10xi64>>,
// CHECK-SAME: %[[VAL_2:.*]]: !fir.ref<!fir.array<10xf32>>) {
// CHECK: %[[VAL_3:.*]] = arith.constant 10 : index
// CHECK: %[[VAL_4:.*]] = arith.constant 100 : index
// CHECK: %[[VAL_5:.*]] = fir.shape %[[VAL_4]] : (index) -> !fir.shape<1>
// CHECK: %[[VAL_6:.*]]:2 = hlfir.declare %[[VAL_0]](%[[VAL_5]]) {uniq_name = "_QFsimpleEx"} : (!fir.ref<!fir.array<100xf32>>, !fir.shape<1>) -> (!fir.ref<!fir.array<100xf32>>, !fir.ref<!fir.array<100xf32>>)
// CHECK: %[[VAL_7:.*]] = fir.shape %[[VAL_3]] : (index) -> !fir.shape<1>
// CHECK: %[[VAL_8:.*]]:2 = hlfir.declare %[[VAL_1]](%[[VAL_7]]) {uniq_name = "y"} : (!fir.ref<!fir.array<10xi64>>, !fir.shape<1>) -> (!fir.ref<!fir.array<10xi64>>, !fir.ref<!fir.array<10xi64>>)
// CHECK: %[[VAL_9:.*]]:2 = hlfir.declare %[[VAL_2]](%[[VAL_7]]) {uniq_name = "z"} : (!fir.ref<!fir.array<10xf32>>, !fir.shape<1>) -> (!fir.ref<!fir.array<10xf32>>, !fir.ref<!fir.array<10xf32>>)
// CHECK: %[[VAL_10:.*]] = arith.constant 1 : index
// CHECK: fir.do_loop %[[VAL_11:.*]] = %[[VAL_10]] to %[[VAL_3]] step %[[VAL_10]] unordered {
// CHECK: %[[VAL_12:.*]] = hlfir.designate %[[VAL_8]]#0 (%[[VAL_11]]) : (!fir.ref<!fir.array<10xi64>>, index) -> !fir.ref<i64>
// CHECK: %[[VAL_13:.*]] = fir.load %[[VAL_12]] : !fir.ref<i64>
// CHECK: %[[VAL_14:.*]] = hlfir.designate %[[VAL_6]]#0 (%[[VAL_13]]) : (!fir.ref<!fir.array<100xf32>>, i64) -> !fir.ref<f32>
// CHECK: %[[VAL_15:.*]] = hlfir.designate %[[VAL_9]]#0 (%[[VAL_11]]) : (!fir.ref<!fir.array<10xf32>>, index) -> !fir.ref<f32>
// CHECK: hlfir.assign %[[VAL_15]] to %[[VAL_14]] : !fir.ref<f32>, !fir.ref<f32>
// CHECK: }
// CHECK: return
// CHECK: }