了解gcc 4.9.2自动矢量化输出 [英] Understanding gcc 4.9.2 auto-vectorization output
问题描述
我正在学习gcc自动矢量化模块。阅读此处的文档。
这是我试过的(debian jessie amd64):
$ cat ex1.c
)。请参阅此处和< a href =https://gcc.gnu.org/onlinedocs/gcc/Debugging-Options.html =nofollow noreferrer>这里:
int a [256],b [256],c [256];
foo(){
int i; (i = 0; i <256; i ++){
a [i] = b [i] + c [i]的
;
$ b然后,我只需运行:$ gcc -xc -Ofast -msse2 -c -ftree -vectorize -fopt -info-vec-missed ex1.c
ex1 .c:5:3:注:misalign = ref字节b [i_11]
ex1.c:5:3:注释:misalign = 0字节ref [i_11]
ex1.c :5:3:note:misalign = 0字节的ref a [i_11]
ex1.c:5:3:注意:虚拟phi。跳跃。
ex1.c:5:3:note:num。 args = 4(不是一元/二元/三元操作)。
ex1.c:5:3:注意:不是ssa-name。
ex1.c:5:3:注意:使用不简单。
ex1.c:5:3:note:num。 args = 4(不是一元/二元/三元操作)。
ex1.c:5:3:注意:不是ssa-name。
ex1.c:5:3:注意:使用不简单。
ex1.c:2:1:注释:未矢量化:基本块中没有足够的数据引用。
ex1.c:6:13:note:not vectorized:no vectype for stmt:vect__4.5_1 = MEM [(int *)vectp_b.3_9];
scalar_type:vector(4)int
ex1.c:6:13:注意:未矢量化:基本块中没有足够的数据引用。
ex1.c:2:1:注释:未矢量化:基本块中没有足够的数据引用。
ex1.c:8:1:注释:未矢量化:基本块中没有足够的数据引用。
根据 documentation ,我会假设看到一条清晰的线条,如下所示:
ex1.c:5:note:LOOP VECTORIZED。
但事实并非如此。我已经使用了命令行选项:
-fopt-info-vec-missed
,因为命令行选项:-ftree-vectorizer-verbose
现在无法使用,如每份报告。
所以我的问题是:我如何阅读上面的输出来提取,以某种方式实际上向量化了循环?
如果有帮助:
$ gcc -dumpversion
4.9.2
解决方案实际上在gcc在线文档中挖掘,我终于发现应该使用
-fopt-info-vec-optimized
(或者-fopt-info-vec-all
< blockquote>
优化
:
优化成功应用时打印信息。决定哪些信息是相关的需要通过。例如,矢量化器通过打印成功向量化的循环的源位置。
错过
:
打印关于错过优化的信息。个人通行证控制哪些信息包含在输出中。
$ b
note
:
打印有关优化的详细信息,例如某些转换,有关决策的更详细信息等等。
全部
:
打印详细的优化信息。这包括'优化','错过'和'注意'。
I am trying to learn gcc auto-vectorization module. After reading documentation from here.
Here is what I tried (debian jessie amd64):
$ cat ex1.c int a[256], b[256], c[256]; foo () { int i; for (i=0; i<256; i++){ a[i] = b[i] + c[i]; } }
And then, I simply run:
$ gcc -x c -Ofast -msse2 -c -ftree-vectorize -fopt-info-vec-missed ex1.c ex1.c:5:3: note: misalign = 0 bytes of ref b[i_11] ex1.c:5:3: note: misalign = 0 bytes of ref c[i_11] ex1.c:5:3: note: misalign = 0 bytes of ref a[i_11] ex1.c:5:3: note: virtual phi. skip. ex1.c:5:3: note: num. args = 4 (not unary/binary/ternary op). ex1.c:5:3: note: not ssa-name. ex1.c:5:3: note: use not simple. ex1.c:5:3: note: num. args = 4 (not unary/binary/ternary op). ex1.c:5:3: note: not ssa-name. ex1.c:5:3: note: use not simple. ex1.c:2:1: note: not vectorized: not enough data-refs in basic block. ex1.c:6:13: note: not vectorized: no vectype for stmt: vect__4.5_1 = MEM[(int *)vectp_b.3_9]; scalar_type: vector(4) int ex1.c:6:13: note: not vectorized: not enough data-refs in basic block. ex1.c:2:1: note: not vectorized: not enough data-refs in basic block. ex1.c:8:1: note: not vectorized: not enough data-refs in basic block.
As per the documentation, I would have assumed to see a clear line saying something like:
ex1.c:5: note: LOOP VECTORIZED.
but that is not the case. I have used the command line option:
-fopt-info-vec-missed
since the command line option:-ftree-vectorizer-verbose
is now a no-op, as per report.So my question: is how do I read the above output to extract that somehow the loop was actually vectorized ?
In case that help:
$ gcc -dumpversion 4.9.2
解决方案Actually digging in gcc online doc, I finally found out that I should use instead:
-fopt-info-vec-optimized
(or maybe-fopt-info-vec-all
). See here and here:
optimized
: Print information when an optimization is successfully applied. It is up to a pass to decide which information is relevant. For example, the vectorizer passes print the source location of loops which are successfully vectorized.
missed
: Print information about missed optimizations. Individual passes control which information to include in the output.
note
: Print verbose information about optimizations, such as certain transformations, more detailed messages about decisions etc.
all
: Print detailed optimization information. This includes ‘optimized’, ‘missed’, and ‘note’.
这篇关于了解gcc 4.9.2自动矢量化输出的文章就介绍到这了,希望我们推荐的答案对大家有所帮助,也希望大家多多支持IT屋!