Skip to content

AVX512FP16: Adjust builtin for mask complex fma

Current mask/mask3 implementation for complex fma contains duplicated parameter in macro, which may cause error at -O0. Refactor macro implementation to builtins to avoid potential error.

For round intrinsic with NO_ROUND as input, ix86_erase_embedded_rounding erases embedded_rounding upspec but could break other emit_insn in expanders. skip those expanders with multiple emit_insn for this function and check rounding in expander with subst.

gcc/ChangeLog:

* config/i386/avx512fp16intrin.h (_mm512_mask_fcmadd_pch):
Adjust builtin call.
(_mm512_mask3_fcmadd_pch): Likewise.
(_mm512_mask_fmadd_pch): Likewise
(_mm512_mask3_fmadd_pch): Likewise
(_mm512_mask_fcmadd_round_pch): Likewise
(_mm512_mask3_fcmadd_round_pch): Likewise
(_mm512_mask_fmadd_round_pch): Likewise
(_mm512_mask3_fmadd_round_pch): Likewise
(_mm_mask_fcmadd_sch): Likewise
(_mm_mask3_fcmadd_sch): Likewise
(_mm_mask_fmadd_sch): Likewise
(_mm_mask3_fmadd_sch): Likewise
(_mm_mask_fcmadd_round_sch): Likewise
(_mm_mask3_fcmadd_round_sch): Likewise
(_mm_mask_fmadd_round_sch): Likewise
(_mm_mask3_fmadd_round_sch): Likewise
(_mm_fcmadd_round_sch): Likewise
* config/i386/avx512fp16vlintrin.h (_mm_mask_fmadd_pch):
Adjust builtin call.
(_mm_mask3_fmadd_pch): Likewise
(_mm256_mask_fmadd_pch): Likewise
(_mm256_mask3_fmadd_pch): Likewise
(_mm_mask_fcmadd_pch): Likewise
(_mm_mask3_fcmadd_pch): Likewise
(_mm256_mask_fcmadd_pch): Likewise
(_mm256_mask3_fcmadd_pch): Likewise
* config/i386/i386-builtin.def: Add mask3 builtin for complex
fma, and adjust mask_builtin to corresponding expander.
* config/i386/i386-expand.c (ix86_expand_round_builtin):
Skip eraseing embedded rounding for expanders that emits
multiple insns.
* config/i386/sse.md (complexmove): New mode_attr.
(<avx512>_fmaddc_<mode>_mask1<round_expand_name>): New expander.
(<avx512>_fcmaddc_<mode>_mask1<round_expand_name>): Likewise.
(avx512fp16_fmaddcsh_v8hf_mask1<round_expand_name>): Likewise.
(avx512fp16_fcmaddcsh_v8hf_mask1<round_expand_name>): Likewise.
(avx512fp16_fcmaddcsh_v8hf_mask3<round_expand_name>): Likewise.
(avx512fp16_fmaddcsh_v8hf_mask3<round_expand_name>): Likewise.
* config/i386/subst.md (round_embedded_complex): New subst.

gcc/testsuite/ChangeLog:

* gcc.target/i386/avx-1.c: Add new mask3 builtins.
* gcc.target/i386/sse-13.c: Ditto.
* gcc.target/i386/sse-23.c: Ditto.

Merge request reports