[llvm] [GlobaISel] Allow expanding of sdiv -> mul by constant (PR #146504)

[David Green via llvm-commits]

================
@@ -229,12 +229,15 @@ define <4 x i32> @test_bit_sink_operand(<4 x i32> %src, <4 x i32> %dst, <4 x i32
 ; CHECK-GI:       // %bb.0: // %entry
 ; CHECK-GI-NEXT:    sub sp, sp, #32
 ; CHECK-GI-NEXT:    .cfi_def_cfa_offset 32
-; CHECK-GI-NEXT:    asr w9, w0, #31
+; CHECK-GI-NEXT:    // kill: def $w0 killed $w0 def $x0
+; CHECK-GI-NEXT:    sxtw x9, w0
 ; CHECK-GI-NEXT:    mov w8, wzr
 ; CHECK-GI-NEXT:    add x10, sp, #16
 ; CHECK-GI-NEXT:    mov x11, sp
-; CHECK-GI-NEXT:    add w9, w0, w9, lsr #31
-; CHECK-GI-NEXT:    asr w9, w9, #1
+; CHECK-GI-NEXT:    sub x9, x9, x9, lsl #31
+; CHECK-GI-NEXT:    asr x9, x9, #32
+; CHECK-GI-NEXT:    add w9, w9, w0
+; CHECK-GI-NEXT:    add w9, w9, w9, lsr #31
----------------
davemgreen wrote:

Do you know why sdiv by power-2 is no longer as efficient as it was? It can be performed more efficiently by a cmp+add+sel+asr.

https://github.com/llvm/llvm-project/pull/146504