[Bug tree-optimization/115777] [12/13/14/15 regression] Severe performance regression on insertion sort at -O2 or above

[sjames at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115777

--- Comment #1 from Sam James <sjames at gcc dot gnu.org> ---
11 vs 12:
```
--- /dev/fd/63  2024-07-03 22:38:09.929726424 +0100
+++ /dev/fd/62  2024-07-03 22:38:09.932726456 +0100
@@ -5,24 +5,28 @@
 insertion_sort.constprop.0:
 .LFB25:
        .cfi_startproc
-       leaq    4(%rdi), %r8
+       movq    %rdi, %r8
        movl    $1, %esi
        .p2align 4,,10
        .p2align 3
 .L2:
-       movq    %r8, %rax
+       movq    %r8, %rdx
+       jmp     .L3
        .p2align 4,,10
        .p2align 3
+.L5:
+       vmovq   %xmm1, (%rdx)
+       leaq    -4(%rdx), %rax
+       cmpq    %rdi, %rdx
+       je      .L6
+       movq    %rax, %rdx
 .L3:
-       movl    -4(%rax), %edx
-       movl    (%rax), %ecx
-       cmpl    %edx, %ecx
-       jnb     .L6
-       movl    %ecx, -4(%rax)
-       subq    $4, %rax
-       movl    %edx, 4(%rax)
-       cmpq    %rax, %rdi
-       jne     .L3
+       vmovq   (%rdx), %xmm0
+       vmovd   %xmm0, %ecx
+       vpextrd $1, %xmm0, %eax
+       vpshufd $225, %xmm0, %xmm1
+       cmpl    %ecx, %eax
+       jb      .L5
 .L6:
        addq    $1, %rsi
        addq    $4, %r8
@@ -77,7 +81,7 @@
        vzeroupper
        call    clock@PLT
        leaq    .LC0(%rip), %rsi
-       movl    $1, %edi
+       movl    $2, %edi
        subq    %rbx, %rax
        movq    %rax, %rdx
        xorl    %eax, %eax
```