Fix dropped alignment operand in byteAddressBufferLoad specialization

[stramit]

Summary

specializeFuncsForBufferLoadArgs lumped kIROp_ByteAddressBufferLoad in with the generic two-operand "element access" instructions (GetElement, StructuredBufferLoad, etc.) in both gatherCallInfoForArg and getSpecializedValueForArg. byteAddressBufferLoad actually has three operands — (buffer, offset, alignment) — so the rewritten load inside the specialized callee was constructed with only two operands.

Downstream, legalizeByteAddressBufferOps::processLoad unconditionally reads operand index 2 to recover the alignment, hitting:

assert failure: slang-ir.h(710): index < getOperandCount()

during SPIR-V emission.

Repro

A struct large enough to trip the buffer-load specialization threshold, loaded via Buf.Load<T>() and passed by value into a helper:

struct Frame
{
    float4x4 a;
    float4x4 b;
    float3 v;
    float pad;
};

float3 makeRay(Frame f)
{
    float4 w = mul(float4(0, 0, 0, 1), f.b);
    return w.xyz + f.v;
}

uniform ByteAddressBuffer buf;
uniform RWTexture2D<float4> outTexture;
uniform uint width;
uniform uint height;

[shader("compute")]
[numthreads(8, 8, 1)]
void computeMain(uint3 dtid : SV_DispatchThreadID)
{
    if (dtid.x >= width || dtid.y >= height) return;
    Frame f = buf.Load<Frame>(0);
    outTexture[dtid.xy] = float4(makeRay(f), 1);
}

slangc repro.slang -target spirv -O0 -fvk-use-scalar-layout aborts with the assertion above.

Fix

Give byteAddressBufferLoad dedicated handling in both functions so the alignment operand survives specialization, and remove it from isElementAccessInst since it doesn't share the two-operand shape.

Test plan

• New regression test tests/optimization/buffer-load-specialize-byte-address.slang exercises the failing pattern on SPIR-V and checks that the specialized helper takes the offset and alignment as separate uint parameters.
• No existing tests change.

---

Generated by Claude Code