[Bug target/114323] [14 Regression] MVE vector load intrinsic miscompiled since r14-5622-g4d7647edfd7d98

[clyon at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114323

Christophe Lyon <clyon at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2024-03-14
             Status|UNCONFIRMED                 |ASSIGNED
     Ever confirmed|0                           |1

--- Comment #2 from Christophe Lyon <clyon at gcc dot gnu.org> ---
Tried a similar testcase for aarch64 (the intrinsics framework for MVE is very
close to SVE's), and noticed that when running dse_classify_store() it sees a
use of the initialized memory:

svuint32_t foo ()  
{
  const uint32_t D.11583[4];
  svuint32_t V0;

<bb 2> : 
 D.11583[0] = 1; 
 D.11583[1] = 2; 
 D.11583[2] = 3; 
 D.11583[3] = 4; 
 V0_7 = svld1_u32 ({ -1, 0, 0, 0, ... }, &D.11583);
 D.11583 ={v} {CLOBBER(eol)};
 return V0_7;
}

# .MEM_6 = VDEF <.MEM_
D.11583[3] = 4;
# VUSE <.MEM_6>
V0_7 = svld1_u32 ({ -1, 0, 0, 0, ... }, &D.11583);



dse_classify_store() on arm/MVE does not see such a use and decides the
initialization is a dead store.