Created attachment 40093 [details] reproducer > g++ -O0 main.cpp func.cpp -o no_opt; ./no_opt 0 > g++ -O1 main.cpp func.cpp -o opt; ./opt 2 > cat main.cpp #include <stdio.h> char C = 0; int I = 197412621; void foo(); int main() { foo(); printf("%d\n", C); } > cat func.cpp extern char C; extern int I; void foo() { C = 0 > (short) (I >> 11); } > g++ -v Using built-in specs. COLLECT_GCC=g++ COLLECT_LTO_WRAPPER=/home/dybaboki/gcc/bin/libexec/gcc/x86_64-pc-linux-gnu/7.0.0/lto-wrapper Target: x86_64-pc-linux-gnu Configured with: ../gcc/configure --with-arch=corei7 --with-cpu=corei7 --enable-clocale=gnu --with-system-zlib --enable-shared --with-demangler-in-ld --enable-cloog-backend=isl --with-fpmath=sse --prefix=/home/dybaboki/gcc/bin --enable-languages=c,c++,fortran,lto : (reconfigured) ../gcc/configure --with-arch=corei7 --with-cpu=corei7 --enable-clocale=gnu --with-system-zlib --enable-shared --with-demangler-in-ld --enable-cloog-backend=isl --with-fpmath=sse --prefix=/home/dybaboki/gcc/bin --enable-languages=c,c++,fortran,lto : (reconfigured) ../gcc/configure --with-arch=corei7 --with-cpu=corei7 --enable-clocale=gnu --with-system-zlib --enable-shared --with-demangler-in-ld --enable-cloog-backend=isl --with-fpmath=sse --prefix=/home/dybaboki/gcc/bin --enable-languages=c,c++,fortran,lto : (reconfigured) ../gcc_github/configure --with-arch=corei7 --with-cpu=corei7 --enable-clocale=gnu --with-system-zlib --enable-shared --with-demangler-in-ld --enable-cloog-backend=isl --with-fpmath=sse --prefix=/home/dybaboki/gcc/bin --enable-languages=c,c++,fortran,lto --no-create --no-recursion : (reconfigured) ../gcc/configure --with-arch=corei7 --with-cpu=corei7 --enable-clocale=gnu --with-system-zlib --enable-shared --with-demangler-in-ld --enable-cloog-backend=isl --with-fpmath=sse --prefix=/home/dybaboki/gcc/bin --enable-languages=c,c++,fortran,lto --no-create --no-recursion Thread model: posix gcc version 7.0.0 20161119 (experimental) (GCC)
Started with r242414. Simplified testcase: char a = 0; int b = 197412621; __attribute__ ((noinline, noclone)) void foo () { a = 0 > (short) (b >> 11); } int main () { asm volatile ("" : : : "memory"); foo (); if (a != 0) __builtin_abort (); return 0; }
Before combine we have: (insn 6 5 7 2 (parallel [ (set (reg:SI 92) (ashiftrt:SI (reg:SI 93 [ b ]) (const_int 11 [0xb]))) (clobber (reg:CC 17 flags)) ]) "pr78438.c":7 563 {*ashrsi3_1} (expr_list:REG_DEAD (reg:SI 93 [ b ]) (expr_list:REG_UNUSED (reg:CC 17 flags) (expr_list:REG_EQUAL (ashiftrt:SI (mem/c:SI (symbol_ref:DI ("b") [flags 0x2] <var_decl 0x7f8ffbb66ea0 b>) [1 b+0 S4 A32]) (const_int 11 [0xb])) (nil))))) (insn 7 6 8 2 (parallel [ (set (reg:HI 94) (lshiftrt:HI (subreg:HI (reg:SI 92) 0) (const_int 15 [0xf]))) (clobber (reg:CC 17 flags)) ]) "pr78438.c":7 572 {*lshrhi3_1} (expr_list:REG_DEAD (reg:SI 92) (expr_list:REG_UNUSED (reg:CC 17 flags) (nil)))) combine turns that into non-equivalent: (insn 7 6 8 2 (parallel [ (set (reg:SI 94) (ashiftrt:SI (reg:SI 93 [ b ]) (const_int 26 [0x1a]))) (clobber (reg:CC 17 flags)) ]) "pr78438.c":7 563 {*ashrsi3_1} (expr_list:REG_UNUSED (reg:CC 17 flags) (expr_list:REG_DEAD (reg:SI 93 [ b ]) (nil))))
The over-sized bitfield isn't really needed for this, making it more severe: struct S { long int : 23; long int a : 24; long int b : 10; long int c : 24; signed char d : 8; } s; __attribute__((noinline, noclone)) void foo () { s.d = 0; s.c = -1193165L; } int main () { foo (); if (s.d != 0) __builtin_abort (); return 0; }
Oops, #c3 was meant for PR78436.
Why is this wrong? This isn't the same reg 94 (it is reused): Failed to match this instruction: (set (mem/c:QI (symbol_ref:DI ("a") [flags 0x2] <var_decl 0x3fffa1d60cf0 a>) [0 a+0 S1 A8]) (subreg:QI (ashiftrt:SI (reg:SI 93 [ b ]) (const_int 26 [0x1a])) 0)) Successfully matched this instruction: (set (reg:SI 94) (ashiftrt:SI (reg:SI 93 [ b ]) (const_int 26 [0x1a]))) Successfully matched this instruction: (set (mem/c:QI (symbol_ref:DI ("a") [flags 0x2] <var_decl 0x3fffa1d60cf0 a>) [0 a+0 S1 A8]) (subreg:QI (reg:SI 94) 0)) The low 8 bits of r93 a>> 26 are exactly what we need to store afaics?
n/m, too tired I guess, please ignore comment 5.
Author: segher Date: Wed Nov 23 14:33:13 2016 New Revision: 242757 URL: https://gcc.gnu.org/viewcvs?rev=242757&root=gcc&view=rev Log: combine: Convert subreg-of-lshiftrt to zero_extract properly (PR78390) r242414, for PR77881, introduces some bugs (PR78390, PR78438, PR78477). It all has the same root cause: that patch makes combine convert every lowpart subreg of a logical shift right to a zero_extract. This cannot work at all if it is not a constant shift, and it has to be a bit more careful exactly which bits it extracts. PR target/77881 PR bootstrap/78390 PR target/78438 PR bootstrap/78477 * combine.c (make_compound_operation_int): Do not convert a subreg of a non-constant logical shift right to a zero_extract. Handle the case where some zero bits have been shifted into the range covered by that subreg. Modified: trunk/gcc/ChangeLog trunk/gcc/combine.c
Author: jakub Date: Fri Nov 25 19:24:22 2016 New Revision: 242883 URL: https://gcc.gnu.org/viewcvs?rev=242883&root=gcc&view=rev Log: PR rtl-optimization/78438 PR rtl-optimization/78477 * gcc.c-torture/execute/pr78438.c: New test. * gcc.c-torture/execute/pr78477.c: New test. Added: trunk/gcc/testsuite/gcc.c-torture/execute/pr78438.c trunk/gcc/testsuite/gcc.c-torture/execute/pr78477.c Modified: trunk/gcc/testsuite/ChangeLog
Fixed.