Difference between mm512_scalef_ps and mm512_scalef_round_ps

[simba611]

While trying to emulate mm512_scalef_ps and mm512_scalef_round_ps, I noticed that there is no difference in the pseudocode provided in the Intel Intrinsics Guide.
Could it be clarified what the difference is between in the two intrinsics ?

[AdyT_Intel]

The mm512_scalef_ps does the rounding based on the rounding mode in the MXCSR.

The mm512_scalef_round_ps gets an additional rounding control parameter and use this rounding instead of the default (in MXCSR).
You can clearly see this in the description below (taken from the Intel intrinsic guide).

Scale the packed single-precision (32-bit) floating-point elements in a using values from b, and store the results in dst.
Rounding is done according to the rounding[3:0] parameter, which can be one of:

(_MM_FROUND_TO_NEAREST_INT |_MM_FROUND_NO_EXC) // round to nearest, and suppress exceptions (_MM_FROUND_TO_NEG_INF |_MM_FROUND_NO_EXC) // round down, and suppress exceptions (_MM_FROUND_TO_POS_INF |_MM_FROUND_NO_EXC) // round up, and suppress exceptions (_MM_FROUND_TO_ZERO |_MM_FROUND_NO_EXC) // truncate, and suppress exceptions _MM_FROUND_CUR_DIRECTION // use MXCSR.RC; see _MM_SET_ROUNDING_MODE

[nemequ]

The rounding you're talking about is the FLOOR() function in the pseudo-code, or is it something in addition to the FLOOR()?

Assuming the former, instead of FLOOR(tmp_src2[31:0]) shouldn't the pseudo-code be ROUND(tmp_src2[i+31:i], rounding[3:0]) for scalef_round and something like ROUND(tmp_src2[i+31:i], MXCSR[2:0] | _MM_FROUND_NO_EXC) for scalef?

[simba611]

I have since tried using _mm_setcsr to set the rounding mode before calling scalef to emulate scalef_round behaviour.
I was unable to emulate scalef_round using this.