I'd like to update the ML on control-flow progress and demonstrate some problems.

The WIP repo is here: https://github.com/HansKristian-Work/DXIL2SPIRV. I'm developing it as a standalone module for time being.

Control flow in DXIL is a complicated beast as it's a soup of gotos, as it is LLVM. The only saving grace is that it must be reducible, i.e. no branching straight into a loop, or arbitrary backward gotos.

The main problem with emitting SPIR-V is:

- For every conditional branch we need a selection merge construct with a unique merge block which header dominates.
- For every loop header, we need a loop merge with designated continue block and unique merge block which header dominates.
- Cannot break out of more than one loop construct at a time (guess what DXIL does!).

The main complication currently is that we need ladder breaking. Here's a concrete example:

cbuffer Buff : register(b10, space1)
{
�� int count1;
�� int count2;
�� int data[1024];
};

float get_r()
{
�� float r = 0.0;
�� [loop]
�� for (int i = 0; i < count1; i++)
�� {
�� [loop]
�� for (int j = 0; j < count2; j++)
�� {
�� if (data[i ^ j] == 40)
�� return r;�� // <-- goto end; when inlined
�� r += float(data[i ^ j]);
�� }
�� }
�� return r;
}

float4 main(float4 pos : POSITION, float4 pos2 : COLOR) : SV_Position
{
�� float r = get_r();
�� return r.xxxx;
}

This gets compiled into:

...

define void @main() {
�� %Buff_cbuffer = call %dx.types.Handle @dx.op.createHandle(i32 57, i8 2, i32 0, i32 10, i1 false)�� ; CreateHandle(resourceClass,rangeId,index,nonUniformIndex)
�� %1 = call %dx.types.CBufRet.i32 @dx.op.cbufferLoadLegacy.i32(i32 59, %dx.types.Handle %Buff_cbuffer, i32 0)�� ; CBufferLoadLegacy(handle,regIndex)
�� %2 = extractvalue %dx.types.CBufRet.i32 %1, 0
�� %3 = icmp sgt i32 %2, 0
�� br i1 %3, label %.lr.ph2.preheader, label %"\01?get_r@@YAMXZ.exit"

.lr.ph2.preheader:�� ; preds = %0
�� br label %.lr.ph2

.lr.ph2:�� ; preds = %._crit_edge, %.lr.ph2.preheader
�� %i.i.0 = phi i32 [ %19, %._crit_edge ], [ 0, %.lr.ph2.preheader ]
�� %r.i.0 = phi float [ %r.i.2, %._crit_edge ], [ 0.000000e+00, %.lr.ph2.preheader ]
�� %4 = call %dx.types.CBufRet.i32 @dx.op.cbufferLoadLegacy.i32(i32 59, %dx.types.Handle %Buff_cbuffer, i32 0)�� ; CBufferLoadLegacy(handle,regIndex)
�� %5 = extractvalue %dx.types.CBufRet.i32 %4, 1
�� %6 = icmp sgt i32 %5, 0
�� br i1 %6, label %.lr.ph.preheader, label %._crit_edge

.lr.ph.preheader:�� ; preds = %.lr.ph2
�� br label %.lr.ph

.lr.ph:�� ; preds = %12, %.lr.ph.preheader
�� %j.i.0 = phi i32 [ %15, %12 ], [ 0, %.lr.ph.preheader ]
�� %r.i.1 = phi float [ %14, %12 ], [ %r.i.0, %.lr.ph.preheader ]
�� %7 = xor i32 %j.i.0, %i.i.0
�� %8 = add i32 %7, 1
�� %9 = call %dx.types.CBufRet.i32 @dx.op.cbufferLoadLegacy.i32(i32 59, %dx.types.Handle %Buff_cbuffer, i32 %8)�� ; CBufferLoadLegacy(handle,regIndex)
�� %10 = extractvalue %dx.types.CBufRet.i32 %9, 0
�� %11 = icmp eq i32 %10, 40
�� br i1 %11, label %"\01?get_r@@YAMXZ.exit.loopexit", label %12

; <label>:12�� ; preds = %.lr.ph
�� %13 = sitofp i32 %10 to float
�� %14 = fadd fast float %13, %r.i.1
�� %15 = add nuw nsw i32 %j.i.0, 1
�� %16 = call %dx.types.CBufRet.i32 @dx.op.cbufferLoadLegacy.i32(i32 59, %dx.types.Handle %Buff_cbuffer, i32 0)�� ; CBufferLoadLegacy(handle,regIndex)
�� %17 = extractvalue %dx.types.CBufRet.i32 %16, 1
�� %18 = icmp slt i32 %15, %17
�� br i1 %18, label %.lr.ph, label %._crit_edge.loopexit, !llvm.loop !25

._crit_edge.loopexit:�� ; preds = %12
�� br label %._crit_edge

._crit_edge:�� ; preds = %._crit_edge.loopexit, %.lr.ph2
�� %r.i.2 = phi float [ %r.i.0, %.lr.ph2 ], [ %14, %._crit_edge.loopexit ]
�� %19 = add nuw nsw i32 %i.i.0, 1
�� %20 = call %dx.types.CBufRet.i32 @dx.op.cbufferLoadLegacy.i32(i32 59, %dx.types.Handle %Buff_cbuffer, i32 0)�� ; CBufferLoadLegacy(handle,regIndex)
�� %21 = extractvalue %dx.types.CBufRet.i32 %20, 0
�� %22 = icmp slt i32 %19, %21
�� br i1 %22, label %.lr.ph2, label %"\01?get_r@@YAMXZ.exit.loopexit.11", !llvm.loop !27

"\01?get_r@@YAMXZ.exit.loopexit":�� ; preds = %.lr.ph
�� br label %"\01?get_r@@YAMXZ.exit"

"\01?get_r@@YAMXZ.exit.loopexit.11":�� ; preds = %._crit_edge
�� br label %"\01?get_r@@YAMXZ.exit"

"\01?get_r@@YAMXZ.exit":�� ; preds = %"\01?get_r@@YAMXZ.exit.loopexit.11", %"\01?get_r@@YAMXZ.exit.loopexit", %0
�� %.0 = phi float [ 0.000000e+00, %0 ], [ %r.i.1, %"\01?get_r@@YAMXZ.exit.loopexit" ], [ %r.i.2, %"\01?get_r@@YAMXZ.exit.loopexit.11" ]
�� call void @dx.op.storeOutput.f32(i32 5, i32 0, i32 0, i8 0, float %.0)�� ; StoreOutput(outputSigId,rowIndex,colIndex,value)
�� call void @dx.op.storeOutput.f32(i32 5, i32 0, i32 0, i8 1, float %.0)�� ; StoreOutput(outputSigId,rowIndex,colIndex,value)
�� call void @dx.op.storeOutput.f32(i32 5, i32 0, i32 0, i8 2, float %.0)�� ; StoreOutput(outputSigId,rowIndex,colIndex,value)
�� call void @dx.op.storeOutput.f32(i32 5, i32 0, i32 0, i8 3, float %.0)�� ; StoreOutput(outputSigId,rowIndex,colIndex,value)
�� ret void
}

...

The "return r;" inside the inner loop turns into a full "exit". This is a very common pattern where inlined leaf functions return. For a normal HLSL -> SPIR-V path, spirv-opt carefully introduces ladders to convert a return into a chain of breaks. We have no such luxury in DXIL. Basically, the return becomes a forward goto.

After a lot of implementation weirdness, I can turn this into SPIR-V which validates. Only control flow is emitted since I haven't looked at actual codegen. The resulting GLSL from SPIRV-Cross ends up looking something like:

#version 450

void main()
{
�� bool COND1;
�� if (COND1)
�� {
�� bool _lr_ph2_preheader;
�� bool _lr_ph2;
�� bool _lr_ph2_succ;
�� bool _lr_ph_preheader;
�� bool _lr_ph;
�� bool COND11;
�� bool _get_r_YAMXZ_exit_loopexit;
�� bool _crit_edge_loopexit_pred;
�� bool _crit_edge_loopexit;
�� bool _crit_edge;
�� do
�� {
�� if (_lr_ph2_succ)
�� {
�� do
�� {
�� if (_lr_ph)
�� {
�� break; // <-- goto end;
�� }
�� else
�� {
�� }
�� } while (!COND11);
�� if (!_crit_edge_loopexit_pred) // <-- Ladder to handle goto
�� {
�� break;
�� }
�� }
�� } while (!_crit_edge);
�� bool _get_r_YAMXZ_exit_loopexit_11_pred;
�� if (_get_r_YAMXZ_exit_loopexit_11_pred) // <-- Ladder to handle goto
�� {
�� bool _get_r_YAMXZ_exit_loopexit_11;
�� }
�� bool _get_r_YAMXZ_exit_loopexit_11_succ;
�� }
// end:
�� bool _get_r_YAMXZ_exit;
}

Some other complications left to consider will be to deal with Phi nodes properly, and switch blocks.

Cheers,
Hans-Kristian