为什么Clang要优化这段代码？
为什么Clang要优化这段代码？
提问于 2014-05-24 05:36:37
代码的目的是找出代表0到1之间的值的32位浮点位模式的总数。在我看来，这应该是可行的，但出于某种原因，Clang的汇编输出基本上等同于return 0;。
我使用-std=c++1y -Wall -Wextra -pedantic -O2和-std=c++1y -Wall -Wextra -pedantic -O3在Clang3.3和Clang3.4.1上编译了这段代码
Clang3.4使用-O2和-O3优化了一切。
Clang3.3只能通过-O3来优化一切。
所谓“优化一切”，我的意思是这是程序的汇编输出：
main: # @main
xorl %eax, %eax
ret#include <limits>
#include <cstring>
#include <cstdint>
template <class TO, class FROM>
inline TO punning_cast(const FROM &input)
{
TO out;
std::memcpy(&out, &input, sizeof(TO));
return out;
}
int main()
{
uint32_t i = std::numeric_limits<uint32_t>::min();
uint32_t count = 0;
while (1)
{
float n = punning_cast<float>(i);
if(n >= 0.0f && n <= 1.0f)
count++;
if (i == std::numeric_limits<uint32_t>::max())
break;
i++;
}
return count;
}回答 2
Stack Overflow用户
回答已采纳
发布于 2014-05-24 07:08:50
下面是一个更简单的测试用例，它指出这是一个编译器错误：
http://coliru.stacked-crooked.com/a/58b3f9b4edd8e373
#include <cstdint>
int main()
{
uint32_t i = 0;
uint32_t count = 1;
while (1)
{
if( i < 5 )
count+=1;
if (i == 0xFFFFFFFF)
break;
i++;
}
return count; // should return 6
}程序集显示它输出的是1，而不是6。它也不认为这是一个无限循环，在这种情况下，程序集不会从main返回。
Stack Overflow用户
发布于 2014-05-25 08:24:37
使用mukunda上面的例子，在带有-O2的clang3.4中，这个问题似乎处于矢量化阶段。向量化的代码在入口跳过向量化的代码：
br i1 true, label %middle.block, label %vector.ph因此，count的值与其初始化时保持不变。
*** IR Dump Before Combine redundant instructions ***
; Function Attrs: nounwind readnone ssp uwtable
define i32 @main() #0 {
entry:
br i1 true, label %middle.block, label %vector.ph
vector.ph: ; preds = %entry
br label %vector.body
vector.body: ; preds = %vector.body, %vector.ph
%index = phi i32 [ 0, %vector.ph ], [ %index.next, %vector.body ]
%vec.phi = phi <4 x i32> [ <i32 1, i32 0, i32 0, i32 0>, %vector.ph ], [ %4, %vector.body ]
%vec.phi8 = phi <4 x i32> [ zeroinitializer, %vector.ph ], [ %5, %vector.body ]
%broadcast.splatinsert = insertelement <4 x i32> undef, i32 %index, i32 0
%broadcast.splat = shufflevector <4 x i32> %broadcast.splatinsert, <4 x i32> undef, <4 x i32> zeroinitializer
%induction = add <4 x i32> %broadcast.splat, <i32 0, i32 1, i32 2, i32 3>
%induction7 = add <4 x i32> %broadcast.splat, <i32 4, i32 5, i32 6, i32 7>
%0 = icmp ult <4 x i32> %induction, <i32 5, i32 5, i32 5, i32 5>
%1 = icmp ult <4 x i32> %induction7, <i32 5, i32 5, i32 5, i32 5>
%2 = zext <4 x i1> %0 to <4 x i32>
%3 = zext <4 x i1> %1 to <4 x i32>
%4 = add <4 x i32> %2, %vec.phi
%5 = add <4 x i32> %3, %vec.phi8
%6 = icmp eq <4 x i32> %induction, <i32 -1, i32 -1, i32 -1, i32 -1>
%7 = icmp eq <4 x i32> %induction7, <i32 -1, i32 -1, i32 -1, i32 -1>
%8 = add <4 x i32> %induction, <i32 1, i32 1, i32 1, i32 1>
%9 = add <4 x i32> %induction7, <i32 1, i32 1, i32 1, i32 1>
%index.next = add i32 %index, 8
%10 = icmp eq i32 %index.next, 0
br i1 %10, label %middle.block, label %vector.body, !llvm.loop !1
middle.block: ; preds = %vector.body, %entry
%resume.val = phi i32 [ 0, %entry ], [ 0, %vector.body ]
%trunc.resume.val = phi i32 [ 0, %entry ], [ 0, %vector.body ]
%rdx.vec.exit.phi = phi <4 x i32> [ <i32 1, i32 0, i32 0, i32 0>, %entry ], [ %4, %vector.body ]
%rdx.vec.exit.phi9 = phi <4 x i32> [ zeroinitializer, %entry ], [ %5, %vector.body ]
%bin.rdx = add <4 x i32> %rdx.vec.exit.phi9, %rdx.vec.exit.phi
%rdx.shuf = shufflevector <4 x i32> %bin.rdx, <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
%bin.rdx10 = add <4 x i32> %bin.rdx, %rdx.shuf
%rdx.shuf11 = shufflevector <4 x i32> %bin.rdx10, <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
%bin.rdx12 = add <4 x i32> %bin.rdx10, %rdx.shuf11
%11 = extractelement <4 x i32> %bin.rdx12, i32 0
%cmp.n = icmp eq i32 0, %resume.val
br i1 %cmp.n, label %while.end, label %scalar.ph
scalar.ph: ; preds = %middle.block
br label %while.body
while.body: ; preds = %while.body, %scalar.ph
%i.0 = phi i32 [ %trunc.resume.val, %scalar.ph ], [ %inc, %while.body ]
%count.0 = phi i32 [ %11, %scalar.ph ], [ %add.count.0, %while.body ]
%cmp = icmp ult i32 %i.0, 5
%add = zext i1 %cmp to i32
%add.count.0 = add i32 %add, %count.0
%cmp1 = icmp eq i32 %i.0, -1
%inc = add i32 %i.0, 1
br i1 %cmp1, label %while.end, label %while.body, !llvm.loop !4
while.end: ; preds = %middle.block, %while.body
%add.count.0.lcssa = phi i32 [ %add.count.0, %while.body ], [ %11, %middle.block ]
ret i32 %add.count.0.lcssa
}优化器随后会删除无法访问和无效的代码--这些代码几乎是整个函数体。
页面原文内容由Stack Overflow提供。腾讯云小微IT领域专用引擎提供翻译支持
原文链接：
https://stackoverflow.com/questions/23838661
复制相关文章
相似问题