eba35cc1c0
This PR proposes re-modelling `reduce` specifiers to match OpenMP and OpenACC. In particular, this PR includes the following: * A new `fir` op: `fir.delcare_reduction` which is identical to OpenMP's `omp.declare_reduction` op. * Updating the `reduce` clause on `fir.do_concurrent.loop` to use the new op. * Re-uses the `ReductionProcessor` component to emit reductions for `do concurrent` just like we do for OpenMP. To do this, the `ReductionProcessor` had to be refactored to be more generalized. * Upates mapping `do concurrent` to `fir.loop ... unordered` nests using the new reduction model. Unfortunately, this is a big PR that would be difficult to divide up in smaller parts because the bottom of the changes are the `fir` table-gen changes to `do concurrent`. However, doing these MLIR changes cascades to the other parts that have to be modified to not break things. This PR goes in the same direction we went for `private/local` speicifiers. Now the `do concurrent` and OpenMP (and OpenACC) dialects are modelled in essentially the same way which makes mapping between them more trivial, hopefully. PR stack: - https://github.com/llvm/llvm-project/pull/145837 (this one) - https://github.com/llvm/llvm-project/pull/146025 - https://github.com/llvm/llvm-project/pull/146028 - https://github.com/llvm/llvm-project/pull/146033
223 lines
9.8 KiB
Plaintext
223 lines
9.8 KiB
Plaintext
// Test fir.do_concurrent operation parse, verify (no errors), and unparse
|
|
|
|
// RUN: fir-opt %s | fir-opt | FileCheck %s
|
|
|
|
func.func @dc_1d(%i_lb: index, %i_ub: index, %i_st: index) {
|
|
fir.do_concurrent {
|
|
%i = fir.alloca i32
|
|
fir.do_concurrent.loop (%i_iv) = (%i_lb) to (%i_ub) step (%i_st) {
|
|
%0 = fir.convert %i_iv : (index) -> i32
|
|
fir.store %0 to %i : !fir.ref<i32>
|
|
}
|
|
}
|
|
return
|
|
}
|
|
|
|
// CHECK-LABEL: func.func @dc_1d
|
|
// CHECK-SAME: (%[[I_LB:.*]]: index, %[[I_UB:.*]]: index, %[[I_ST:.*]]: index)
|
|
// CHECK: fir.do_concurrent {
|
|
// CHECK: %[[I:.*]] = fir.alloca i32
|
|
// CHECK: fir.do_concurrent.loop (%[[I_IV:.*]]) = (%[[I_LB]]) to (%[[I_UB]]) step (%[[I_ST]]) {
|
|
// CHECK: %[[I_IV_CVT:.*]] = fir.convert %[[I_IV]] : (index) -> i32
|
|
// CHECK: fir.store %[[I_IV_CVT]] to %[[I]] : !fir.ref<i32>
|
|
// CHECK: }
|
|
// CHECK: }
|
|
|
|
func.func @dc_2d(%i_lb: index, %i_ub: index, %i_st: index,
|
|
%j_lb: index, %j_ub: index, %j_st: index) {
|
|
fir.do_concurrent {
|
|
%i = fir.alloca i32
|
|
%j = fir.alloca i32
|
|
fir.do_concurrent.loop
|
|
(%i_iv, %j_iv) = (%i_lb, %j_lb) to (%i_ub, %j_ub) step (%i_st, %j_st) {
|
|
%0 = fir.convert %i_iv : (index) -> i32
|
|
fir.store %0 to %i : !fir.ref<i32>
|
|
|
|
%1 = fir.convert %j_iv : (index) -> i32
|
|
fir.store %1 to %j : !fir.ref<i32>
|
|
}
|
|
}
|
|
return
|
|
}
|
|
|
|
// CHECK-LABEL: func.func @dc_2d
|
|
// CHECK-SAME: (%[[I_LB:.*]]: index, %[[I_UB:.*]]: index, %[[I_ST:.*]]: index, %[[J_LB:.*]]: index, %[[J_UB:.*]]: index, %[[J_ST:.*]]: index)
|
|
// CHECK: fir.do_concurrent {
|
|
// CHECK: %[[I:.*]] = fir.alloca i32
|
|
// CHECK: %[[J:.*]] = fir.alloca i32
|
|
// CHECK: fir.do_concurrent.loop
|
|
// CHECK-SAME: (%[[I_IV:.*]], %[[J_IV:.*]]) = (%[[I_LB]], %[[J_LB]]) to (%[[I_UB]], %[[J_UB]]) step (%[[I_ST]], %[[J_ST]]) {
|
|
// CHECK: %[[I_IV_CVT:.*]] = fir.convert %[[I_IV]] : (index) -> i32
|
|
// CHECK: fir.store %[[I_IV_CVT]] to %[[I]] : !fir.ref<i32>
|
|
// CHECK: %[[J_IV_CVT:.*]] = fir.convert %[[J_IV]] : (index) -> i32
|
|
// CHECK: fir.store %[[J_IV_CVT]] to %[[J]] : !fir.ref<i32>
|
|
// CHECK: }
|
|
// CHECK: }
|
|
|
|
func.func @dc_2d_reduction(%i_lb: index, %i_ub: index, %i_st: index,
|
|
%j_lb: index, %j_ub: index, %j_st: index) {
|
|
%sum = fir.alloca i32
|
|
|
|
fir.do_concurrent {
|
|
%i = fir.alloca i32
|
|
%j = fir.alloca i32
|
|
fir.do_concurrent.loop
|
|
(%i_iv, %j_iv) = (%i_lb, %j_lb) to (%i_ub, %j_ub) step (%i_st, %j_st)
|
|
reduce(@add_reduction_i32 #fir.reduce_attr<add> %sum -> %sum_arg : !fir.ref<i32>) {
|
|
%0 = fir.convert %i_iv : (index) -> i32
|
|
fir.store %0 to %i : !fir.ref<i32>
|
|
|
|
%1 = fir.convert %j_iv : (index) -> i32
|
|
fir.store %1 to %j : !fir.ref<i32>
|
|
}
|
|
}
|
|
return
|
|
}
|
|
|
|
// CHECK-LABEL: func.func @dc_2d_reduction
|
|
// CHECK-SAME: (%[[I_LB:.*]]: index, %[[I_UB:.*]]: index, %[[I_ST:.*]]: index, %[[J_LB:.*]]: index, %[[J_UB:.*]]: index, %[[J_ST:.*]]: index)
|
|
|
|
// CHECK: %[[SUM:.*]] = fir.alloca i32
|
|
|
|
// CHECK: fir.do_concurrent {
|
|
// CHECK: %[[I:.*]] = fir.alloca i32
|
|
// CHECK: %[[J:.*]] = fir.alloca i32
|
|
// CHECK: fir.do_concurrent.loop
|
|
// CHECK-SAME: (%[[I_IV:.*]], %[[J_IV:.*]]) = (%[[I_LB]], %[[J_LB]]) to (%[[I_UB]], %[[J_UB]]) step (%[[I_ST]], %[[J_ST]]) reduce(@add_reduction_i32 #fir.reduce_attr<add> %[[SUM]] -> %{{.*}} : !fir.ref<i32>) {
|
|
// CHECK: %[[I_IV_CVT:.*]] = fir.convert %[[I_IV]] : (index) -> i32
|
|
// CHECK: fir.store %[[I_IV_CVT]] to %[[I]] : !fir.ref<i32>
|
|
// CHECK: %[[J_IV_CVT:.*]] = fir.convert %[[J_IV]] : (index) -> i32
|
|
// CHECK: fir.store %[[J_IV_CVT]] to %[[J]] : !fir.ref<i32>
|
|
// CHECK: }
|
|
// CHECK: }
|
|
|
|
fir.local {type = local} @local_privatizer : i32
|
|
|
|
// CHECK: fir.local {type = local} @[[LOCAL_PRIV_SYM:local_privatizer]] : i32
|
|
|
|
fir.local {type = local_init} @local_init_privatizer : i32 copy {
|
|
^bb0(%arg0: !fir.ref<i32>, %arg1: !fir.ref<i32>):
|
|
%0 = fir.load %arg0 : !fir.ref<i32>
|
|
fir.store %0 to %arg1 : !fir.ref<i32>
|
|
fir.yield(%arg1 : !fir.ref<i32>)
|
|
}
|
|
|
|
// CHECK: fir.local {type = local_init} @[[LOCAL_INIT_PRIV_SYM:local_init_privatizer]] : i32
|
|
// CHECK: ^bb0(%[[ORIG_VAL:.*]]: !fir.ref<i32>, %[[LOCAL_VAL:.*]]: !fir.ref<i32>):
|
|
// CHECK: %[[ORIG_VAL_LD:.*]] = fir.load %[[ORIG_VAL]]
|
|
// CHECK: fir.store %[[ORIG_VAL_LD]] to %[[LOCAL_VAL]] : !fir.ref<i32>
|
|
// CHECK: fir.yield(%[[LOCAL_VAL]] : !fir.ref<i32>)
|
|
// CHECK: }
|
|
|
|
func.func @do_concurrent_with_locality_specs() {
|
|
%3 = fir.alloca i32 {bindc_name = "local_init_var"}
|
|
%4:2 = hlfir.declare %3 {uniq_name = "_QFdo_concurrentElocal_init_var"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
|
|
%5 = fir.alloca i32 {bindc_name = "local_var"}
|
|
%6:2 = hlfir.declare %5 {uniq_name = "_QFdo_concurrentElocal_var"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
|
|
%c1 = arith.constant 1 : index
|
|
%c10 = arith.constant 10 : index
|
|
|
|
fir.do_concurrent {
|
|
%9 = fir.alloca i32 {bindc_name = "i"}
|
|
%10:2 = hlfir.declare %9 {uniq_name = "_QFdo_concurrentEi"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
|
|
|
|
fir.do_concurrent.loop (%arg0) = (%c1) to (%c10) step (%c1)
|
|
local(@local_privatizer %6#0 -> %arg1, @local_init_privatizer %4#0 -> %arg2 : !fir.ref<i32>, !fir.ref<i32>) {
|
|
%11 = fir.convert %arg0 : (index) -> i32
|
|
fir.store %11 to %10#0 : !fir.ref<i32>
|
|
%13:2 = hlfir.declare %arg1 {uniq_name = "_QFdo_concurrentElocal_var"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
|
|
%15:2 = hlfir.declare %arg2 {uniq_name = "_QFdo_concurrentElocal_init_var"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
|
|
}
|
|
}
|
|
return
|
|
}
|
|
|
|
// CHECK-LABEL: func.func @do_concurrent_with_locality_specs() {
|
|
// CHECK: %[[LOC_INIT_ALLOC:.*]] = fir.alloca i32 {bindc_name = "local_init_var"}
|
|
// CHECK: %[[LOC_INIT_DECL:.*]]:2 = hlfir.declare %[[LOC_INIT_ALLOC]]
|
|
|
|
// CHECK: %[[LOC_ALLOC:.*]] = fir.alloca i32 {bindc_name = "local_var"}
|
|
// CHECK: %[[LOC_DECL:.*]]:2 = hlfir.declare %[[LOC_ALLOC]]
|
|
|
|
// CHECK: %[[C1:.*]] = arith.constant 1 : index
|
|
// CHECK: %[[C10:.*]] = arith.constant 10 : index
|
|
|
|
// CHECK: fir.do_concurrent {
|
|
// CHECK: %[[DC_I_ALLOC:.*]] = fir.alloca i32 {bindc_name = "i"}
|
|
// CHECK: %[[DC_I_DECL:.*]]:2 = hlfir.declare %[[DC_I_ALLOC]]
|
|
|
|
// CHECK: fir.do_concurrent.loop (%[[IV:.*]]) = (%[[C1]]) to
|
|
// CHECK-SAME: (%[[C10]]) step (%[[C1]])
|
|
// CHECK-SAME: local(@[[LOCAL_PRIV_SYM]] %[[LOC_DECL]]#0 -> %[[LOC_ARG:[^,]*]],
|
|
// CHECK-SAME: @[[LOCAL_INIT_PRIV_SYM]] %[[LOC_INIT_DECL]]#0 -> %[[LOC_INIT_ARG:.*]] :
|
|
// CHECK-SAME: !fir.ref<i32>, !fir.ref<i32>) {
|
|
|
|
// CHECK: %[[IV_CVT:.*]] = fir.convert %[[IV]] : (index) -> i32
|
|
// CHECK: fir.store %[[IV_CVT]] to %[[DC_I_DECL]]#0 : !fir.ref<i32>
|
|
|
|
// CHECK: %[[LOC_PRIV_DECL:.*]]:2 = hlfir.declare %[[LOC_ARG]]
|
|
// CHECK: %[[LOC_INIT_PRIV_DECL:.*]]:2 = hlfir.declare %[[LOC_INIT_ARG]]
|
|
// CHECK: }
|
|
// CHECK: }
|
|
// CHECK: return
|
|
// CHECK: }
|
|
|
|
func.func @dc_reduce() {
|
|
%3 = fir.alloca i32 {bindc_name = "s", uniq_name = "dc_reduce"}
|
|
%4:2 = hlfir.declare %3 {uniq_name = "dc_reduce"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
|
|
%c1 = arith.constant 1 : index
|
|
fir.do_concurrent {
|
|
fir.do_concurrent.loop (%arg0) = (%c1) to (%c1) step (%c1) reduce(byref @add_reduction_i32 #fir.reduce_attr<add> %4#0 -> %arg1 : !fir.ref<i32>) {
|
|
}
|
|
}
|
|
return
|
|
}
|
|
|
|
// CHECK-LABEL: func.func @dc_reduce() {
|
|
// CHECK: %[[S_ALLOC:.*]] = fir.alloca i32 {bindc_name = "s", uniq_name = "dc_reduce"}
|
|
// CHECK: %[[S_DECL:.*]]:2 = hlfir.declare %[[S_ALLOC]] {uniq_name = "dc_reduce"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
|
|
// CHECK: fir.do_concurrent {
|
|
// CHECK: fir.do_concurrent.loop (%{{.*}}) = (%{{.*}}) to (%{{.*}}) step (%{{.*}}) reduce(byref @add_reduction_i32 #fir.reduce_attr<add> %[[S_DECL]]#0 -> %[[S_ARG:.*]] : !fir.ref<i32>) {
|
|
// CHECK: }
|
|
// CHECK: }
|
|
// CHECK: return
|
|
// CHECK: }
|
|
|
|
func.func @dc_reduce_2() {
|
|
%3 = fir.alloca i32 {bindc_name = "s", uniq_name = "dc_reduce"}
|
|
%4:2 = hlfir.declare %3 {uniq_name = "dc_reduce"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
|
|
|
|
%5 = fir.alloca i32 {bindc_name = "m", uniq_name = "dc_reduce"}
|
|
%6:2 = hlfir.declare %5 {uniq_name = "dc_reduce"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
|
|
|
|
%c1 = arith.constant 1 : index
|
|
|
|
fir.do_concurrent {
|
|
fir.do_concurrent.loop (%arg0) = (%c1) to (%c1) step (%c1)
|
|
reduce(@add_reduction_i32 #fir.reduce_attr<add> %4#0 -> %arg1,
|
|
@mul_reduction_i32 #fir.reduce_attr<multiply> %6#0 -> %arg2
|
|
: !fir.ref<i32>, !fir.ref<i32>) {
|
|
}
|
|
}
|
|
|
|
return
|
|
}
|
|
|
|
// CHECK-LABEL: func.func @dc_reduce_2() {
|
|
// CHECK: %[[S_ALLOC:.*]] = fir.alloca i32 {bindc_name = "s", uniq_name = "dc_reduce"}
|
|
// CHECK: %[[S_DECL:.*]]:2 = hlfir.declare %[[S_ALLOC]] {uniq_name = "dc_reduce"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
|
|
|
|
// CHECK: %[[M_ALLOC:.*]] = fir.alloca i32 {bindc_name = "m", uniq_name = "dc_reduce"}
|
|
// CHECK: %[[M_DECL:.*]]:2 = hlfir.declare %[[M_ALLOC]] {uniq_name = "dc_reduce"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
|
|
// CHECK: fir.do_concurrent {
|
|
// CHECK: fir.do_concurrent.loop (%{{.*}}) = (%{{.*}}) to (%{{.*}}) step (%{{[^[:space:]]+}})
|
|
// CHECK-SAME: reduce(
|
|
// CHECK-SAME: @add_reduction_i32 #fir.reduce_attr<add> %[[S_DECL]]#0 -> %[[S_ARG:[^,]+]],
|
|
// CHECK-SAME: @mul_reduction_i32 #fir.reduce_attr<multiply> %[[M_DECL]]#0 -> %[[M_ARG:[^[:space:]]+]]
|
|
// CHECK-SAME: : !fir.ref<i32>, !fir.ref<i32>) {
|
|
// CHECK: }
|
|
// CHECK: }
|
|
// CHECK: return
|
|
// CHECK: }
|
|
|