From 4513c80d81d5797aba8572ccc899a370596d80ce Mon Sep 17 00:00:00 2001
From: "raphael.bach" <raphael.bach@etu.hesge.ch>
Date: Tue, 19 Jul 2022 02:23:44 +0200
Subject: [PATCH] Add more reduce operations in `fmpi_reduce.fut`
---
src/futhark/fmpi_reduce.fut | 79 ++++++++++++++++++++++++++++++++++---
1 file changed, 73 insertions(+), 6 deletions(-)
diff --git a/src/futhark/fmpi_reduce.fut b/src/futhark/fmpi_reduce.fut
index 0bb889e..7e37430 100644
--- a/src/futhark/fmpi_reduce.fut
+++ b/src/futhark/fmpi_reduce.fut
@@ -1,17 +1,84 @@
-- Default operations for MPI_Reduce()
-- See https://www.open-mpi.org/doc/current/man3/MPI_Reduce.3.php#sect10
-local module type fmpi_reduce_t = {
+local module type fmpi_reduce_bool_t = {
type t
- val sum [n]: [n]t -> t
- val prod [n]: [n]t -> t
+ val sum [n]: [n]t -> i64
+ val and [n]: [n]t -> t
+ val or [n]: [n]t -> t
+ val xor [n]: [n]t -> t
+}
+
+local module type fmpi_reduce_integer_t = {
+ type t
+ val sum [n]: [n]t -> i64
+ val prod [n]: [n]t -> i64
+ val min [n]: [n]t -> t
+ val max [n]: [n]t -> t
+ val and [n]: [n]t -> t
+ val or [n]: [n]t -> t
+ val xor [n]: [n]t -> t
+ val & [n]: [n]t -> t
+ val | [n]: [n]t -> t
+ val ^ [n]: [n]t -> t
+}
+
+local module type fmpi_reduce_float_t = {
+ type t
+ val sum [n]: [n]t -> f64
+ val prod [n]: [n]t -> f64
val min [n]: [n]t -> t
val max [n]: [n]t -> t
+ val and [n]: [n]t -> t
+ val or [n]: [n]t -> t
+ val xor [n]: [n]t -> t
}
-module fmpi_reduce (T: float): (fmpi_reduce_t with t = T.t) = {
+module fmpi_reduce_bool: (fmpi_reduce_bool_t with t = bool) = {
+ type t = bool
+ def sum [n] (xs: [n]t): i64 = reduce (+) 0 (map (\x -> i64.bool x) xs)
+ def and [n] (xs: [n]t): t = reduce (&&) true xs
+ def or [n] (xs: [n]t): t = reduce (||) false xs
+ def xor [n] (xs: [n]t): t = reduce (!=) false xs
+}
+
+module fmpi_reduce_integer (T: integral): (fmpi_reduce_integer_t with t = T.t) = {
+ type t = T.t
+ local def and_impl (a: t) (b: t): t = T.bool (a T.!= (T.i32 0) && b T.!= (T.i32 0))
+ local def or_impl (a: t) (b: t): t = T.bool (a T.!= (T.i32 0) || b T.!= (T.i32 0))
+ local def xor_impl (a: t) (b: t): t = T.bool (a T.!= b)
+ def sum [n] (xs: [n]t): i64 = T.to_i64 (T.sum xs)
+ def prod [n] (xs: [n]t): i64 = T.to_i64 (T.product xs)
+ def min [n] (xs: [n]t): t = T.minimum xs
+ def max [n] (xs: [n]t): t = T.maximum xs
+ def and [n] (xs: [n]t): t = reduce (and_impl) (T.i32 1) xs
+ def or [n] (xs: [n]t): t = reduce (or_impl) (T.i32 0) xs
+ def xor [n] (xs: [n]t): t = reduce (xor_impl) (T.i32 0) xs
+ def (&) [n] (xs: [n]t): t = reduce (T.&) (T.i32 1) xs
+ def (|) [n] (xs: [n]t): t = reduce (T.|) (T.i32 0) xs
+ def (^) [n] (xs: [n]t): t = reduce (T.^) (T.i32 0) xs
+}
+
+module fmpi_reduce_float (T: float): (fmpi_reduce_float_t with t = T.t) = {
type t = T.t
- def sum [n] (xs: [n]t): t = T.sum xs
- def prod [n] (xs: [n]t): t = T.product xs
+ local def and_impl (a: t) (b: t): t = T.bool (a T.!= (T.i32 0) && b T.!= (T.i32 0))
+ local def or_impl (a: t) (b: t): t = T.bool (a T.!= (T.i32 0) || b T.!= (T.i32 0))
+ local def xor_impl (a: t) (b: t): t = T.bool (a T.!= b)
+ def sum [n] (xs: [n]t): f64 = T.to_f64 (T.sum xs)
+ def prod [n] (xs: [n]t): f64 = T.to_f64 (T.product xs)
def min [n] (xs: [n]t): t = T.minimum xs
def max [n] (xs: [n]t): t = T.maximum xs
+ def and [n] (xs: [n]t): t = reduce (and_impl) (T.i32 1) xs
+ def or [n] (xs: [n]t): t = reduce (or_impl) (T.i32 0) xs
+ def xor [n] (xs: [n]t): t = reduce (xor_impl) (T.i32 0) xs
}
+
+module fmpi_reduce_i8 = fmpi_reduce_integer i8
+module fmpi_reduce_i16 = fmpi_reduce_integer i16
+module fmpi_reduce_i32 = fmpi_reduce_integer i32
+module fmpi_reduce_i64 = fmpi_reduce_integer i64
+module fmpi_reduce_u8 = fmpi_reduce_integer u8
+module fmpi_reduce_u16 = fmpi_reduce_integer u16
+module fmpi_reduce_u32 = fmpi_reduce_integer u32
+module fmpi_reduce_u64 = fmpi_reduce_integer u64
+module fmpi_reduce_f32 = fmpi_reduce_float f32
+module fmpi_reduce_f64 = fmpi_reduce_float f64
--
GitLab