I was compiling this to a ps_1_4 target:
sampler sampShadowContribution : register( s0 );
sampler sampCurrentFrame : register( s1 );
sampler sampPreviousFrames : register( s2 );
float4 AdditiveCombine( in float2 t : TEXCOORD ) : COLOR
float4 shadow = tex2D( sampShadowContribution, t );
float4 current = tex2D( sampCurrentFrame, t );
float4 previous = tex2D( sampPreviousFrames, t );
return ( shadow * current + previous );
Thus the outputs get clamped to [-1,+1], thus no HDRI data was emitted by the additive combiner. I'll have to do some research, but it would appear that ATI were more leniant and allowed values outside that range. For some reason, [-8,+8] range and ATI sound familiar...
I was damn sure that I was using SM2 (and higher) exclusively throughout this project - but the above sucker was the first shader I implemented, and buried away deep in my codebase. It's also the only "stand alone" pixel shader - so when I checked all my effects files for SM2 paths I overlooked this one.
Back to some Direct3D 10 R&D [grin]