These are the same:

//ORIGINAL

mul r2.xyzw, r0.zzzz, cb1[5].xyzw

r2.x = cb1[5].x * r0.z
r2.y = cb1[5].y * r0.z
r2.z = cb1[5].z * r0.z
r2.w = cb1[5].w * r0.z

mad r2.xyzw, r0.yyyy, cb1[4].xyzw, r2.xyzw

r2.x = cb1[4].x * r0.y + cb1[5].x * r0.z
r2.y = cb1[4].y * r0.y + cb1[5].y * r0.z
r2.z = cb1[4].z * r0.y + cb1[5].z * r0.z
r2.w = cb1[4].w * r0.y + cb1[5].w * r0.z

mad r2.xyzw, r0.xxxx, cb1[6].xyzw, r2.xyzw

r2.x = cb1[6].x * r0.x + cb1[4].x * r0.y + cb1[5].x * r0.z
r2.y = cb1[6].y * r0.x + cb1[4].y * r0.y + cb1[5].y * r0.z
r2.z = cb1[6].z * r0.x + cb1[4].z * r0.y + cb1[5].z * r0.z
r2.w = cb1[6].w * r0.x + cb1[4].w * r0.y + cb1[5].w * r0.z

add r2.xyzw, r2.xywz, cb1[7].xywz

r2.x = cb1[7].x + cb1[6].x * r0.x + cb1[4].x * r0.y + cb1[5].x * r0.z
r2.y = cb1[7].y + cb1[6].y * r0.x + cb1[4].y * r0.y + cb1[5].y * r0.z
r2.z = cb1[7].w + cb1[6].w * r0.x + cb1[4].w * r0.y + cb1[5].w * r0.z
r2.w = cb1[7].z + cb1[6].z * r0.x + cb1[4].z * r0.y + cb1[5].z * r0.z

//RECOMPILED

mul r2.xyzw, r0.zzzz, cb1[5].xywz

r2.x = cb1[5].x * r0.z
r2.y = cb1[5].y * r0.z
r2.z = cb1[5].w * r0.z
r2.w = cb1[5].z * r0.z

mad r2.xyzw, r0.yyyy, cb1[4].xywz, r2.xyzw

r2.x = cb1[4].x * r0.y + cb1[5].x * r0.z
r2.y = cb1[4].y * r0.y + cb1[5].y * r0.z
r2.z = cb1[4].w * r0.y + cb1[5].w * r0.z
r2.w = cb1[4].z * r0.y + cb1[5].z * r0.z

mad r2.xyzw, r0.xxxx, cb1[6].xywz, r2.xyzw

r2.x = cb1[6].x * r0.x + cb1[4].x * r0.y + cb1[5].x * r0.z
r2.y = cb1[6].y * r0.x + cb1[4].y * r0.y + cb1[5].y * r0.z
r2.z = cb1[6].w * r0.x + cb1[4].w * r0.y + cb1[5].w * r0.z
r2.w = cb1[6].z * r0.x + cb1[4].z * r0.y + cb1[5].z * r0.z

add r2.xyzw, r2.xyzw, cb1[7].xywz

r2.x = cb1[7].x + cb1[6].x * r0.x + cb1[4].x * r0.y + cb1[5].x * r0.z
r2.y = cb1[7].y + cb1[6].y * r0.x + cb1[4].y * r0.y + cb1[5].y * r0.z
r2.z = cb1[7].w + cb1[6].w * r0.x + cb1[4].w * r0.y + cb1[5].w * r0.z
r2.w = cb1[7].z + cb1[6].z * r0.x + cb1[4].z * r0.y + cb1[5].z * r0.z

//COMPARISON
*Original:
r2.x = cb1[7].x + cb1[6].x * r0.x + cb1[4].x * r0.y + cb1[5].x * r0.z
r2.y = cb1[7].y + cb1[6].y * r0.x + cb1[4].y * r0.y + cb1[5].y * r0.z
r2.z = cb1[7].w + cb1[6].w * r0.x + cb1[4].w * r0.y + cb1[5].w * r0.z
r2.w = cb1[7].z + cb1[6].z * r0.x + cb1[4].z * r0.y + cb1[5].z * r0.z

• Re-Compiled
  r2.x = cb1[7].x + cb1[6].x * r0.x + cb1[4].x * r0.y + cb1[5].x * r0.z
  r2.y = cb1[7].y + cb1[6].y * r0.x + cb1[4].y * r0.y + cb1[5].y * r0.z
  r2.z = cb1[7].w + cb1[6].w * r0.x + cb1[4].w * r0.y + cb1[5].w * r0.z
  r2.w = cb1[7].z + cb1[6].z * r0.x + cb1[4].z * r0.y + cb1[5].z * r0.z

These are the same.

from 3dmigoto.

Just saw this again recently in Black Ops 3. That swap of of the .xyzw to .xywz I think is an fxc compiler bug, I don't think it's legal.

from 3dmigoto.

Did it actually cause a problem in Black Ops 3? That looks perfectly legal if the compiler decided to store the coordinate in r2.xywz, which it is perfectly legal to do since that is effectively just four independent temporary registers and it can choose whatever order it likes. If it screwed up something in an output register that would be a totally different matter since the order is defined for those.

Regardless, this doesn't look like a 3DMigoto bug - can we close this?

from 3dmigoto.

Closing this due to lack of response.

from 3dmigoto.