diff options
| -rw-r--r-- | erts/doc/src/atomics.xml | 2 | ||||
| -rw-r--r-- | erts/doc/src/counters.xml | 8 | ||||
| -rw-r--r-- | erts/emulator/beam/big.c | 12 | ||||
| -rw-r--r-- | erts/emulator/test/big_SUITE.erl | 18 |
4 files changed, 27 insertions, 13 deletions
diff --git a/erts/doc/src/atomics.xml b/erts/doc/src/atomics.xml index 46231d9234..f552c11e18 100644 --- a/erts/doc/src/atomics.xml +++ b/erts/doc/src/atomics.xml @@ -30,7 +30,7 @@ mutable atomic variables. The implementation utilizes only atomic hardware instructions without any software level locking, which makes it very efficient for concurrent access. The atomics are organized into - arrays with the follwing semantics:</p> + arrays with the following semantics:</p> <list type="bulleted"> <item> <p>Atomics are 64 bit integers.</p> diff --git a/erts/doc/src/counters.xml b/erts/doc/src/counters.xml index c3b0bfcf85..3d26093a59 100644 --- a/erts/doc/src/counters.xml +++ b/erts/doc/src/counters.xml @@ -29,7 +29,7 @@ <p>This module provides a set of functions to do operations towards shared mutable counter variables. The implementation does not utilize any software level locking, which makes it very efficient for concurrent - access. The counters are organized into arrays with the follwing + access. The counters are organized into arrays with the following semantics:</p> <list type="bulleted"> <item> @@ -80,7 +80,7 @@ <taglist> <tag><c>atomics</c> (Default)</tag> <item><p>Counters will be sequentially consistent. If write - operation A is done sequencially before write operation B, then a concurrent reader + operation A is done sequentially before write operation B, then a concurrent reader may see none of them, only A, or both A and B. It cannot see only B.</p> </item> <tag><c>write_concurrency</c></tag> @@ -90,7 +90,7 @@ inconsistency and memory consumption per counter.</p> <p>Read operations may see sequentially inconsistent results with regard to concurrent write operations. Even if write operation A is done - sequencially before write operation B, a concurrent reader may see any + sequentially before write operation B, a concurrent reader may see any combination of A and B, including only B. A read operation is only guaranteed to see all writes done sequentially before the read. No writes are ever lost, but will eventually all be seen.</p> @@ -140,7 +140,7 @@ <c><anno>Ix</anno></c>.</p> <note> <p>Despite its name, the <c>write_concurrency</c> optimization does not - improve <c>put</c>. A call to <c>put</c> is a relative heavy + improve <c>put</c>. A call to <c>put</c> is a relatively heavy operation compared to the very lightweight and scalable <seealso marker="#add/3"><c>add</c></seealso> and <seealso marker="#sub/3"> <c>sub</c></seealso>. The cost for a <c>put</c> with diff --git a/erts/emulator/beam/big.c b/erts/emulator/beam/big.c index 84338769e0..dac9574fa5 100644 --- a/erts/emulator/beam/big.c +++ b/erts/emulator/beam/big.c @@ -668,27 +668,25 @@ static dsize_t I_mul(ErtsDigit* x, dsize_t xl, ErtsDigit* y, dsize_t yl, ErtsDig static dsize_t I_sqr(ErtsDigit* x, dsize_t xl, ErtsDigit* r) { - ErtsDigit d_next = *x; ErtsDigit d; ErtsDigit* r0 = r; ErtsDigit* s = r; if ((r + xl) == x) /* "Inline" operation */ *x = 0; - x++; while(xl--) { - ErtsDigit* y = x; + ErtsDigit* y; ErtsDigit y_0 = 0, y_1 = 0, y_2 = 0, y_3 = 0; ErtsDigit b0, b1; ErtsDigit z0, z1, z2; ErtsDigit t; dsize_t y_l = xl; - + + d = *x; + x++; + y = x; s = r; - d = d_next; - d_next = *x; - x++; DMUL(d, d, b1, b0); DSUMc(*s, b0, y_3, t); diff --git a/erts/emulator/test/big_SUITE.erl b/erts/emulator/test/big_SUITE.erl index 0a42b09903..5b602dd4dc 100644 --- a/erts/emulator/test/big_SUITE.erl +++ b/erts/emulator/test/big_SUITE.erl @@ -168,7 +168,11 @@ eval({op,_,Op,A0,B0}, LFH) -> Res = eval_op(Op, A, B), erlang:garbage_collect(), Res; -eval({integer,_,I}, _) -> I; +eval({integer,_,I}, _) -> + %% "Parasitic" ("symbiotic"?) test of squaring all numbers + %% found in the test data. + test_squaring(I), + I; eval({call,_,{atom,_,Local},Args0}, LFH) -> Args = eval_list(Args0, LFH), LFH(Local, Args). @@ -192,6 +196,18 @@ eval_op('bxor', A, B) -> A bxor B; eval_op('bsl', A, B) -> A bsl B; eval_op('bsr', A, B) -> A bsr B. +test_squaring(I) -> + %% Multiplying an integer by itself is specially optimized, so we + %% should take special care to test squaring. The optimization + %% will kick in when the two operands have the same address. + Sqr = I * I, + + %% This expression will be multiplied in the usual way, because + %% the the two operands for '*' are stored at different addresses. + Sqr = I * ((I + id(1)) - id(1)), + + ok. + %% Built in test functions fac(0) -> 1; |