[Bug c/108593] New: No inlining after function cloning
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108593
Bug ID: 108593
Summary: No inlining after function cloning
Product: gcc
Version: 13.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c
Assignee: unassigned at gcc dot gnu.org
Reporter: david.bolvansky at gmail dot com
Target Milestone: ---
int __attribute__ ((noinline))
foo (int arg)
{
return 2 * arg;
}
int
bar (int arg)
{
return foo (5);
}
results in:
foo.constprop.0:
mov eax, 10
ret
foo:
lea eax, [rdi+rdi]
ret
bar:
jmp foo.constprop.0
But ... why foo.constprop.0 is not inlined fully into bar? Maybe
foo.constprop.0 inherits noinline attribute from foo?
If so, gcc should drop attributes from cloned functions..
[Bug rtl-optimization/7061] Access of bytes in struct parameters
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=7061 Dávid Bolvanský changed: What|Removed |Added CC||david.bolvansky at gmail dot com --- Comment #10 from Dávid Bolvanský --- llvm emits just: im: # @im shufps xmm0, xmm0, 85 # xmm0 = xmm0[1,1,1,1] ret
[Bug ipa/104187] Call site specific attribute to control inliner
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104187
--- Comment #8 from Dávid Bolvanský ---
So this works in Clang now
int foo(int x, int y) { // any compiler will happily inline this function
return x / y;
}
int test(int x, int y) {
int r = 0;
[[clang::noinline]] r += foo(x, y); // for some reason we don't want any
inlining here
return r;
}
foo(int, int): # @foo(int, int)
mov eax, edi
cdq
idiv esi
ret
test(int, int): # @test(int, int)
jmp foo(int, int) # TAILCALL
foo(int, int): # @foo(int, int)
mov eax, edi
cdq
idiv esi
ret
test(int, int): # @test(int, int)
jmp foo(int, int) # TAILCALL
[Bug ipa/104187] Call site specific attribute to control inliner
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104187 --- Comment #5 from Dávid Bolvanský --- So you prefer eg. g = a[i] - [[gnu::always_inline]] foo(x, y) + 2 * bar(); over g = a[i] - __builtin_always_inline(foo(x, y)) + 2 * bar(); ? What is your proposed syntax?
[Bug c/104187] New: Call site specific attribute to control inliner
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104187
Bug ID: 104187
Summary: Call site specific attribute to control inliner
Product: gcc
Version: unknown
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c
Assignee: unassigned at gcc dot gnu.org
Reporter: david.bolvansky at gmail dot com
Target Milestone: ---
It could be useful to have more control over inlining. Use cases:
int foo();
void bar();
int g;
void test()
{
g = __builtin_always_inline(foo()); // force inlining of foo() here
__builtin_noinline(bar()); // never inline bar to this function
}
[Bug tree-optimization/93150] (A) == CST1 &( ((A)==CST2) | ((A)==CST3) ) is not simplified
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=93150 Dávid Bolvanský changed: What|Removed |Added CC||david.bolvansky at gmail dot com --- Comment #2 from Dávid Bolvanský --- Bin ops with constants are simplified by compiler itself..
[Bug tree-optimization/103002] New: Missed loop unrolling opportunity
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103002
Bug ID: 103002
Summary: Missed loop unrolling opportunity
Product: gcc
Version: unknown
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: david.bolvansky at gmail dot com
Target Milestone: ---
#define C 3
struct node {
struct node *next;
int payload;
};
static int count_nodes(const node* p) {
int size = 0;
while (p) {
p = p->next;
size++;
}
return size;
}
bool has_one_node(const node* p) {
return count_nodes(p) == 1;
}
bool has_C_nodes(const node* p) {
return count_nodes(p) == C;
}
has_one_node(node const*):# @has_one_node(node const*)
testrdi, rdi
je .LBB0_1
mov eax, 1
.LBB0_3:# =>This Inner Loop Header: Depth=1
mov rdi, qword ptr [rdi]
add eax, -1
testrdi, rdi
jne .LBB0_3
testeax, eax
seteal
ret
.LBB0_1:
xor eax, eax
ret
has_C_nodes(node const*): # @has_C_nodes(node const*)
testrdi, rdi
je .LBB1_1
mov eax, 3
.LBB1_3:# =>This Inner Loop Header: Depth=1
mov rdi, qword ptr [rdi]
add eax, -1
testrdi, rdi
jne .LBB1_3
testeax, eax
seteal
ret
.LBB1_1:
xor eax, eax
ret
has_C_nodes is simple with some kind of loop deletion pass, but generally,
these loops can be unrolled for some reasonable C values.
https://godbolt.org/z/do656c17b
[Bug tree-optimization/102564] New: Missed loop vectorization with reduction and ptr load/store inside loop
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102564
Bug ID: 102564
Summary: Missed loop vectorization with reduction and ptr
load/store inside loop
Product: gcc
Version: unknown
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: david.bolvansky at gmail dot com
Target Milestone: ---
void test1(int *p, int *t, int N) {
for (int i = 0; i != N; i++) *t += p[i];
}
void test2(int *p, int *t, int N) {
if (N > 1024) // hint, N is not small
for (int i = 0; i != N; i++) *t += p[i];
}
void test3(int *p, int *t, int N) {
if (N > 1024) { // hint, N is not small
int s = 0;
for (int i = 0; i != N; i++) s += p[i];
*t += s;
}
}
test3 is successfully vectorized with LLVM, GCC, ICC. Sadly, only ICC can catch
test1 and test2.
https://godbolt.org/z/PzoYd4eEK
[Bug tree-optimization/102483] New: Regression in codegen of reduction of 4 chars
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102483
Bug ID: 102483
Summary: Regression in codegen of reduction of 4 chars
Product: gcc
Version: unknown
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: david.bolvansky at gmail dot com
Target Milestone: ---
char foo (char* p)
{
char sum = 0;
for (int i = 0; i != 4; i++)
sum += p[i];
return sum;
}
-O3 -march=x86-64
GCC trunk:
foo:
mov edx, DWORD PTR [rdi]
movzx eax, dh
mov ecx, edx
add eax, edx
shr ecx, 16
add eax, ecx
shr edx, 24
add eax, edx
ret
GCC 11 (much better):
foo:
movzx eax, BYTE PTR [rdi+1]
add al, BYTE PTR [rdi]
add al, BYTE PTR [rdi+2]
add al, BYTE PTR [rdi+3]
ret
Best? llvm-mca says so..
foo:# @foo
movdxmm0, dword ptr [rdi] # xmm0 = mem[0],zero,zero,zero
pxorxmm1, xmm1
psadbw xmm1, xmm0
movdeax, xmm1
ret
https://godbolt.org/z/sT9svvj7W
[Bug c/100260] New: DSE: join stores
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100260
Bug ID: 100260
Summary: DSE: join stores
Product: gcc
Version: tree-ssa
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c
Assignee: unassigned at gcc dot gnu.org
Reporter: david.bolvansky at gmail dot com
Target Milestone: ---
#include
struct pam {
void *p1;
void *p2;
#ifdef LONG
unsigned long size;
#else
unsigned int pad;
unsigned int size;
#endif
};
extern int use(struct pam *param);
unsigned int foo(void) {
struct pam s_pam;
memset(_pam, 0, sizeof(struct pam));
s_pam.size = 1;
return use(_pam);
}
INT
foo():
sub rsp, 40
pxor xmm0, xmm0
mov rdi, rsp
mov DWORD PTR [rsp+16], 0
mov DWORD PTR [rsp+20], 1
movaps XMMWORD PTR [rsp], xmm0
call use(pam*)
add rsp, 40
ret
LONG
foo():
sub rsp, 40
pxor xmm0, xmm0
mov rdi, rsp
movaps XMMWORD PTR [rsp], xmm0
mov QWORD PTR [rsp+16], 1
call use(pam*)
add rsp, 40
ret
Stores
mov DWORD PTR [rsp+16], 0
mov DWORD PTR [rsp+20], 1
can be replaced with one mov QWORD..
[Bug tree-optimization/99971] GCC generates partially vectorized and scalar code at once
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99971 Dávid Bolvanský changed: What|Removed |Added CC||david.bolvansky at gmail dot com --- Comment #7 from Dávid Bolvanský --- Still bad for -O3 -march=skylake-avx512 https://godbolt.org/z/azb8aTG43
[Bug target/98348] [10 Regression] GCC 10.2 AVX512 Mask regression from GCC 9
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98348 Dávid Bolvanský changed: What|Removed |Added CC||david.bolvansky at gmail dot com --- Comment #20 from Dávid Bolvanský --- Some small regression (missed opportunity to use vptestnmd): Current trunk compare(unsigned int __vector(16)): vpxor xmm1, xmm1, xmm1 vpcmpd k0, zmm0, zmm1, 0 vpmovm2d zmm0, k0 ret GCC 9.2 compare(unsigned int __vector(16)): vptestnmd k0, zmm0, zmm0 vpmovm2d zmm0, k0 ret https://gcc.godbolt.org/z/5vK68jM3r
[Bug middle-end/98713] Failure to generate branch version of abs if user requested it
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98713 --- Comment #5 from Dávid Bolvanský --- User knows the data better, so he/she may prefer abs with branch. Also PGO may say that branch for abs is better based on profile data.
[Bug c/98713] New: Failure to generate branch version of abs if user requested it
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98713
Bug ID: 98713
Summary: Failure to generate branch version of abs if user
requested it
Product: gcc
Version: 11.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c
Assignee: unassigned at gcc dot gnu.org
Reporter: david.bolvansky at gmail dot com
Target Milestone: ---
int branch_abs(int v) {
return __builtin_expect(v > 0, 1) ? v : -v;
}
GCC -O2 now:
branch_abs:
mov eax, edi
neg eax
cmovs eax, edi
ret
Expected:
branch_abs:
mov eax, edi
test edi, edi
js .LBB0_1
ret
.LBB0_1:
neg eax
ret
Same for min/max.
[Bug other/98663] gcc generates endless loop at -O2 or greater depending on order of testExpression
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98663 Dávid Bolvanský changed: What|Removed |Added CC||david.bolvansky at gmail dot com --- Comment #1 from Dávid Bolvanský --- Compiler can do anything if there is UB in the code.
[Bug c/98658] Loop idiom recognization for memcpy/memmove
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98658 --- Comment #3 from Dávid Bolvanský --- Yes, runtime check.
[Bug c/98658] Loop idiom recognization for memcpy/memmove
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98658 --- Comment #1 from Dávid Bolvanský --- ICC produces memcpy: https://godbolt.org/z/oKxxTM
[Bug c/98658] New: Loop idiom recognization for memcpy/memmove
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98658
Bug ID: 98658
Summary: Loop idiom recognization for memcpy/memmove
Product: gcc
Version: 11.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c
Assignee: unassigned at gcc dot gnu.org
Reporter: david.bolvansky at gmail dot com
Target Milestone: ---
void copy(int *__restrict__ d, int * s, __SIZE_TYPE__ sz) {
__SIZE_TYPE__ i;
for (i = 0; i < sz; i++) {
*d++ = *s++;
}
}
gcc emits call to memcpy.
void copy(int * d, int * s, __SIZE_TYPE__ sz) {
__SIZE_TYPE__ i;
for (i = 0; i < sz; i++) {
*d++ = *s++;
}
}
gcc could emit memmove, but currently does not:
https://godbolt.org/z/5n1rnh
