When an integer is converted into floating point in subword vector extract, it can be done in 2 instructions instead of the 3+ instructions it generates right now. This patch removes the uncessary generation. Differential: https://reviews.llvm.org/D100604
5.6 KiB
5.6 KiB