* Improvements and fixes in Keras loss masking:
    * Whether you represent a ragged tensor as a `tf.RaggedTensor` or using [keras masking](https://www.tensorflow.org/guide/keras/masking_and_padding), the returned loss values should be the identical to each other. In previous versions Keras may have silently ignored the mask.

On POWER8/ppc64le, emulate the POWER9 instructions by loading unaligned // doublewords and byte-swapping each doubleword to emulate BE load/stores. #define NEEDS_ESPERM #define P8_LXVB16X(RA,RB,VT) \ LXVD2X (RA+RB), VT \ VPERM VT, VT, ESPERM, VT #define P8_STXVB16X(VS,RA,RB) \ VPERM VS, VS, ESPERM, TMP2 \ STXVD2X TMP2, (RA+RB) #define XXBRD_ON_LE(VA,VT) \ VPERM VA, VA, ESPERM, VT // Setup byte-swapping permute value in ESPERM for POWER9 instruction // emulation macros. #define SETUP_ESPERM(rtmp) \...

On POWER8/ppc64le, emulate the POWER9 instructions by loading unaligned // doublewords and byte-swapping each doubleword to emulate BE load/stores. #define NEEDS_ESPERM #define P8_LXVB16X(RA,RB,VT) \ LXVD2X (RA+RB), VT \ VPERM VT, VT, ESPERM, VT #define P8_STXVB16X(VS,RA,RB) \ VPERM VS, VS, ESPERM, TMP2 \ STXVD2X TMP2, (RA+RB) #define XXBRD_ON_LE(VA,VT) \ VPERM VA, VA, ESPERM, VT // Setup byte-swapping permute value in ESPERM for POWER9 instruction // emulation macros. #define SETUP_ESPERM(rtmp) \...