[Bug target/90878] [8/9/10 Regression] integer -> SSE register move isn't generated

[hjl.tools at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90878

--- Comment #1 from H.J. Lu <hjl.tools at gmail dot com> ---
If we make integer register store more expensive, this testcase will
regress:

[hjl@gnu-cfl-1 unroll]$ cat x.i
void
foo (long p2, long *diag, long d, long i)
{
  long k;
  k = p2 < 3 ? p2 + p2 : p2 + 3;
  while (i < k)
    diag[i++] = d;
}
[hjl@gnu-cfl-1 unroll]$ make
/export/build/gnu/tools-build/gcc-wip-debug/build-x86_64-linux/gcc/xgcc
-B/export/build/gnu/tools-build/gcc-wip-debug/build-x86_64-linux/gcc/ -O3
-march=skylake  -S x.i
[hjl@gnu-cfl-1 unroll]$ cat x.s
        .file   "x.i"
        .text
        .p2align 4
        .globl  foo
        .type   foo, @function
foo:
.LFB0:
        .cfi_startproc
        leaq    (%rdi,%rdi), %rax
        leaq    3(%rdi), %r8
        cmpq    $2, %rdi
        cmovle  %rax, %r8
        cmpq    %rcx, %r8
        jle     .L10
        movq    %rcx, %rax
        notq    %rax
        movq    %r8, %r9
        addq    %r8, %rax
        subq    %rcx, %r9
        cmpq    $3, %rax
        jbe     .L5
        movq    %r9, %rdi
        shrq    $2, %rdi
        vmovq   %rdx, %xmm1
        leaq    (%rsi,%rcx,8), %rax
        salq    $5, %rdi
        vpbroadcastq    %xmm1, %ymm0
        addq    %rax, %rdi
        .p2align 4,,10
        .p2align 3
.L6:
        vmovdqu %ymm0, (%rax)
        addq    $32, %rax
        cmpq    %rdi, %rax
        jne     .L6
        movq    %r9, %rax
        andq    $-4, %rax
        addq    %rax, %rcx
        cmpq    %rax, %r9
        je      .L9
        vzeroupper
.L5:
        leaq    1(%rcx), %rax
        movq    %rdx, (%rsi,%rcx,8)
        cmpq    %r8, %rax
        jge     .L10
        leaq    2(%rcx), %rdi
        movq    %rdx, (%rsi,%rax,8)
        cmpq    %rdi, %r8
        jle     .L10
        addq    $3, %rcx
        movq    %rdx, (%rsi,%rdi,8)
        cmpq    %rcx, %r8
        jle     .L10
        movq    %rdx, (%rsi,%rcx,8)
        ret
        .p2align 4,,10
        .p2align 3
.L9:
        vzeroupper
.L10:
        ret
        .cfi_endproc
.LFE0:
        .size   foo, .-foo
        .ident  "GCC: (GNU) 10.0.0 20190613 (experimental)"
        .section        .note.GNU-stack,"",@progbits
[hjl@gnu-cfl-1 unroll]$ 

since higher integer register store cost will reduce loop unroll count.