Optimizer incorrectly transforms sdiv to udiv when divisor is shifted INT_MIN

Description

struct UniformBuffer {
    uint a;          // Value: 31
};

struct StorageBuffer {
    int b;
};

cbuffer u_input : register(b0) { UniformBuffer u_input; }
RWByteAddressBuffer s_output : register(u1);

[numthreads(1, 1, 1)]
void main()
{
    // Should be 10 / (INT_MIN >> 31) = 10 / -1 = -10
    s_output.Store(0, int(10) / (int(-2147483647 - 1) >> u_input.a));
}

Expected Behavior
Given u_input.a = 31, the shader should output -10.

Actual Behavior
At -O0, the output is -10 (correct).
At -O1 and higher, the output is 0.

Steps to Reproduce
https://shader-playground.timjones.io/2ffb96015bb368393bb19608833d7d87
The difference between -O0 and -O1 seems to be this:

At -O0,
  %7 = ashr i32 -2147483648, %6
- %8 = sdiv i32 10, %7

At -O1,
  %7 = ashr i32 -2147483648, %6
+ %8 = udiv i32 10, %7

Environment

• DXC version dxcompiler.dll: 1.10 - 1.9.0.15173 (main, d6b78b745); dxil.dll: 1.10(1.9.0.15173)
• Host Operating System Microsoft Windows 11 Home,10.0.26200 N/A Build 26200

Start with the linked Shader Playground reproduction and compare the generated IR at -O0 with -O1, focusing on the sdiv-to-udiv transformation after the arithmetic shift. Done means optimization preserves signed division for this shifted INT_MIN case and the shader produces -10 instead of 0.