Prevent rtl optimization of vec_duplicate + zero_extend to vpbroadcastm since there could be an extra kmov after RA.
Bootstrapped and regtested on x86_64-pc-linux-gnu{-m32,} Ready to push to trunk. gcc/ChangeLog: PR target/110788 * config/i386/sse.md (avx512cd_maskb_vec_dup<mode>): Add UNSPEC_MASKOP. (avx512cd_maskw_vec_dup<mode>: Ditto. gcc/testsuite/ChangeLog: * gcc.target/i386/pr110788.c: New test. --- gcc/config/i386/sse.md | 8 ++++++-- gcc/testsuite/gcc.target/i386/pr110788.c | 11 +++++++++++ 2 files changed, 17 insertions(+), 2 deletions(-) create mode 100644 gcc/testsuite/gcc.target/i386/pr110788.c diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md index 35fd66ed4aa..51961bbfc0b 100644 --- a/gcc/config/i386/sse.md +++ b/gcc/config/i386/sse.md @@ -26778,11 +26778,14 @@ (define_insn "<mask_codefor>avx512dq_broadcast<mode><mask_name>_1" (set_attr "prefix" "evex") (set_attr "mode" "<sseinsnmode>")]) +;; Use unspec to prevent rtl optimizer to optimize zero_extend + vec_duplicate +;; to pbroadcastm, there could be an extra kmov after RA. (define_insn "avx512cd_maskb_vec_dup<mode>" [(set (match_operand:VI8_AVX512VL 0 "register_operand" "=v") (vec_duplicate:VI8_AVX512VL (zero_extend:DI - (match_operand:QI 1 "register_operand" "k"))))] + (match_operand:QI 1 "register_operand" "k")))) + (unspec [(const_int 0)] UNSPEC_MASKOP)] "TARGET_AVX512CD" "vpbroadcastmb2q\t{%1, %0|%0, %1}" [(set_attr "type" "mskmov") @@ -26793,7 +26796,8 @@ (define_insn "avx512cd_maskw_vec_dup<mode>" [(set (match_operand:VI4_AVX512VL 0 "register_operand" "=v") (vec_duplicate:VI4_AVX512VL (zero_extend:SI - (match_operand:HI 1 "register_operand" "k"))))] + (match_operand:HI 1 "register_operand" "k")))) + (unspec [(const_int 0)] UNSPEC_MASKOP)] "TARGET_AVX512CD" "vpbroadcastmw2d\t{%1, %0|%0, %1}" [(set_attr "type" "mskmov") diff --git a/gcc/testsuite/gcc.target/i386/pr110788.c b/gcc/testsuite/gcc.target/i386/pr110788.c new file mode 100644 index 00000000000..4cf1676ccb6 --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/pr110788.c @@ -0,0 +1,11 @@ +/* { dg-do compile } */ +/* { dg-options "-O3 -march=cascadelake --param vect-partial-vector-usage=2" } */ +/* { dg-final { scan-assembler-not "vpbroadcastm" } } */ + +double a[1024], b[1024]; + +void foo (int n) +{ + for (int i = 0; i < n; ++i) + a[i] = b[i] * 3.; +} -- 2.39.1.388.g2fc9e9ca3c