Building under ubuntu 12.04.

[billyevans]

Hi. Thank you for your tool. That's pretty cool.
Your last commit kills speed and building under ubuntu 12.04.

I have tested your previous version and it compressed my 3.3 GB file in 2.6 sec.
But last has 16.35 sec and that time worse than time of pure lz4 library.
And now it doesn't eat 600-800% of cpu just only 150%.

I made a little patch for building under precise. But I am not sure it's nice.
gcc 4.6 doesn't have support for std::launch::deferred.

diff --git a/src/lz4mt.cpp b/src/lz4mt.cpp
index 49c1775..a5a19f4 100644
--- a/src/lz4mt.cpp
+++ b/src/lz4mt.cpp
@@ -8,6 +8,13 @@
 #include "lz4mt_xxh32.h"
 #include "lz4mt_mempool.h"

+#define GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__)
+
+#if (GCC_VERSION == 406)
+#define LAUNCH_SINGLE_THREAD (std::launch::sync)
+#else
+#define LAUNCH_SINGLE_THREAD (std::launch::deferred)
+#endif

 namespace {

@@ -339,7 +346,7 @@ lz4mtCompress(Lz4MtContext* ctx, const Lz4MtStreamDescriptor* sd)
        const bool singleThread      = 0 != (ctx->mode & LZ4MT_MODE_SEQUENTIAL);
        const auto nConcurrency      = std::thread::hardware_concurrency();
        const auto nPool             = singleThread ? 1 : nConcurrency + 1;
-       const auto launch            = singleThread ? std::launch::deferred : std::launch::async;
+       const auto launch            = singleThread ? LAUNCH_SINGLE_THREAD : std::launch::async;

        Lz4Mt::MemPool srcBufferPool(nBlockMaximumSize, nPool);
        Lz4Mt::MemPool dstBufferPool(nBlockMaximumSize, nPool);
@@ -535,7 +542,7 @@ lz4mtDecompress(Lz4MtContext* ctx, Lz4MtStreamDescriptor* sd)
                const bool singleThread      = 0 != (ctx->mode & LZ4MT_MODE_SEQUENTIAL);
                const auto nConcurrency      = std::thread::hardware_concurrency();
                const auto nPool             = singleThread ? 1 : nConcurrency + 1;
-               const auto launch            = singleThread ? std::launch::deferred : std::launch::async;
+               const auto launch            = singleThread ? LAUNCH_SINGLE_THREAD : std::launch::async;

                Lz4Mt::MemPool srcBufferPool(nBlockMaximumSize, nPool);
                Lz4Mt::MemPool dstBufferPool(nBlockMaximumSize, nPool);

[t-mat]

Thanks for the report and patch.
I forgot about gcc 4.6 completely 😨

The patch looks good. I'll add #ifdef __GNUC__ for MSVC++ and merge it after some testing.

[t-mat]

I've investigated, and I realised there are two problems.

• As Alexey mentioned, there is no std::launch::deferred.
• Also std::thread::hardware_concurrency() always returns 0.

Commit 6863cb0 for issue/4 branch will fix these problems.

Please try it.

[billyevans]

Thank you.
Everything work fine. And it works faster! 1.75782sec. And doesn't eat a lot of memory while decompressing.

My little advice for you: add flags -Wall -W -Werror -Weffc++ -Wextra in Makefile. It can be usefull.

This one -Weffc++ is a little annoying, but sometimes it can help with uninitialized fields.

[t-mat]

Thanks for testing and advice !
I'll check your compile flags.