The eliminate reg-reg move by inverting the condition of
a cmove #2 peephole2 converts the following sequence:
473: bx:DI=[r14:DI*0x8+r12:DI]
960: r15:DI=r8:DI
485: {flags:CCC=cmp(r15:DI+bx:DI,bx:DI);r15:DI=r15:DI+bx:DI;}
737: r15:DI={(geu(flags:CCC,0))?r15:DI:bx:DI}
to:
1110: {flags:CCC=cmp(r8:DI+bx:DI,bx:DI);r8:DI=r8:DI+bx:DI;}
1111: r15:DI=[r14:DI*0x8+r12:DI]
1112: r15:DI={(geu(flags:CCC,0))?r8:DI:r15:DI}
Please note that(insn 1110) uses register BX, but its
initialization was eliminated.
Avoid conversion if eliminated move intialized a register, used
in the moved instruction.
2022-11-03 Uroš Bizjak <ubiz...@gmail.com>
gcc/ChangeLog:
PR target/107404
* config/i386/i386.md (eliminate reg-reg move by inverting the
condition of a cmove #2 peephole2): Check if eliminated move
initialized a register, used in the moved instruction.
gcc/testsuite/ChangeLog:
PR target/107404
* g++.target/i386/pr107404.C: New test.
Bootstrapped and regression tested on x86_64-linux-gnu {,-m32}.
Pushed to master.
Uros.
diff --git a/gcc/config/i386/i386.md b/gcc/config/i386/i386.md
index 85567980aa3..436eabb691a 100644
--- a/gcc/config/i386/i386.md
+++ b/gcc/config/i386/i386.md
@@ -21800,7 +21800,8 @@ (define_peephole2
&& REGNO (operands[2]) != REGNO (operands[1])
&& peep2_reg_dead_p (2, operands[1])
&& peep2_reg_dead_p (4, operands[2])
- && !reg_overlap_mentioned_p (operands[0], operands[3])"
+ && !reg_overlap_mentioned_p (operands[0], operands[3])
+ && !reg_mentioned_p (operands[2], operands[6])"
[(parallel [(set (match_dup 7) (match_dup 8))
(set (match_dup 1) (match_dup 9))])
(set (match_dup 0) (match_dup 3))
diff --git a/gcc/testsuite/g++.target/i386/pr107404.C
b/gcc/testsuite/g++.target/i386/pr107404.C
new file mode 100644
index 00000000000..e47d0fd779d
--- /dev/null
+++ b/gcc/testsuite/g++.target/i386/pr107404.C
@@ -0,0 +1,53 @@
+// PR target/107404
+// { dg-do run }
+// { dg-options "-O3" }
+
+unsigned long long a;
+void b(unsigned long long *f, int p2) { *f ^= p2; }
+long c;
+char e, i;
+short g, m;
+long long ab[1][25][21][22];
+unsigned long long aa[1][21][22];
+unsigned long long ae[1][25][21][21];
+long long ac[129360];
+char ad[25][1][21];
+char ah[1][25][1][21];
+short af[100];
+long max(long f, unsigned long p2) { return f < p2 ? p2 : f; }
+const int &max2(const int &f, const int &p2) { return f < p2 ? p2 : f; }
+void foo(unsigned f, unsigned p2, char l, char p4, long long n[][25][21][22],
+ unsigned long long p6[][21][22], unsigned long long u[][25][21][21]) {
+ long an;
+ for (int j = 0; j < 4; j = p2)
+ for (short k = 0; k < 7; k += 2)
+ for (short o = 0; o < (short)p2 + 21742; o = l) {
+ for (signed char p = 2; p < 9; p += p4)
+ if (p6[j][o][p])
+ for (long q(3); 4 ? n[0][k][o][0] : 0;
+ q += p6[0][o][0] ? p6[j][0][p] : 0)
+ ac[j + q] = 5066799590;
+ for (long r(p4 - 16); r < 21; r += 4) {
+ ad[k][o][r] = max(u[j][k][o][r], f + u[j][k][o][r]);
+ long d = u[j][k][o][r];
+ an = d < p2 ? p2 : d;
+ e = ah[j][k][o][r] = an;
+ af[o * r] = i;
+ }
+ for (short s(c); s < (short)p2; s = 2)
+ for (short am(m); am; am = max2(3, p2))
+ for (long y = 0; y; y = 3)
+ for (short t(0); t < max2(g, 0);)
+ ;
+ }
+}
+int main() {
+ foo(7, 1558227751, 104, 16, ab, aa, ae);
+ for (unsigned long v = 0; v < 5; ++v)
+ for (unsigned long w = 0; w < 1; ++w)
+ for (unsigned long x = 0; x < 21; ++x)
+ b(&a, ad[v][w][x]);
+
+ if (a)
+ __builtin_abort();
+}
