[hipblaslt] Fix fails with dtl.yaml and xfp32.yaml on gfx950_mx_rebase

[nakajee]

Motivation

Fix fails with dtl.yaml and xfp32.yaml on gfx950_mx_rebase branch

Technical Details

• Fixed merge issue with if kernel["ScheduleIterAlg"] == 3
• Added int cast for float value const in asm
• Fixed incorrect parameter for calcLdsBlockSizePerPad()
• Fixed incorrect local read calculation due to incorrectly applying MX logic to TF32 emulation
• Fixed incorrect ShiftK code vreg due to missing if condition for TF32 emulation

Test Plan

ran dtl.yaml and xfp32.yaml on gfx950_mx_rebase branch

Test Result

All passed except for a known issue (should be fixed with latest develop branch)

Submission Checklist

• Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.

[b-shi]

LGTM, thanks for the fix!