[SPIR-V] Embedding debug info from shader source with non Latin letters causes exception

[LawG4]

Description
Attempting to generate spirv byte code with debug information inside the binary crashes when the input file path contains non-latin text. Resulting in an internal compiler error to be reported. Except when specifying vulkan-with-source

Steps to Reproduce

# Crash
dxc.exe hlsl_頂点.hlsl -spirv -T vs_6_0 -Qembed_debug -Zi

# No Crash, no spirv flag set
dxc.exe hlsl_頂点.hlsl -T vs_6_0 -Qembed_debug -Zi

# Crash 
dxc.exe hlsl_頂点.hlsl -spirv -T vs_6_0 -fspv-debug=file
dxc.exe hlsl_頂点.hlsl -spirv -T vs_6_0 -fspv-debug=line
dxc.exe hlsl_頂点.hlsl -spirv -T vs_6_0 -fspv-debug=source

# Interestingly,  this results in *no* crash 
dxc.exe hlsl_頂点.hlsl -spirv -T vs_6_0 -fspv-debug=vulkan-with-source

Source code inside the hlsl file is a simple text shader which just passes information to next stage

struct VS_OUTPUT {
    float4 pos : POSITION;
};

struct VS_INPUT {
    float4 pos : POSITION;
};

VS_OUTPUT main(const VS_INPUT vs_input) {
    VS_OUTPUT vs_out;
    vs_out.pos = vs_input.pos;
    return vs_out;
}

Actual Behavior
The compiler only outputs this text on the runs which crash

Internal compiler error: access violation. Attempted to read from address 0x00000000000000A8

Environment

• DXC version: dxcompiler.dll: 1.8 - 1.8.2407.7 (416fab6); dxil.dll: 1.8(101.8.2407.12)
• Host Operating System: Windows 10

[s-perron]

I do not think this is spirv specific. When I run with a debug build, I see the same assert when targeting DXIL and SPIRV. I'll ask someone at Microsoft to double check.

DirectXShaderCompiler/issues/6948 » dxc hlsl_頂点.hlsl -spirv -T vs_6_0 -Qembed_debug -Zi
otherwise contents changedError: assert(false && "otherwise contents changed")
File:
DirectXShaderCompiler/lib/DxcSupport/Unicode.cpp(192)
Func:	UTF8ToWideString
[1]    513981 illegal hardware instruction  dxc hlsl_頂点.hlsl -spirv -T vs_6_0 -Qembed_debug -Zi
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
DirectXShaderCompiler/issues/6948 » dxc hlsl_頂点.hlsl -T vs_6_0 -Qembed_debug -Zi
otherwise contents changedError: assert(false && "otherwise contents changed")
File:
DirectXShaderCompiler/lib/DxcSupport/Unicode.cpp(192)
Func:	UTF8ToWideString
[1]    514042 illegal hardware instruction  dxc hlsl_頂点.hlsl -T vs_6_0 -Qembed_debug -Zi
-------------------------------------------------------------------------------------------------------

[damyanp]

Thanks - unfortunately, due to our current set of priorities we're going to be unable to look into this issue. We'd happily look at any submitted PRs addressing it though.