https://gcc.gnu.org/g:11ef3c2e4fb97fc200edb74ce602960af0eaa354
commit r16-7098-g11ef3c2e4fb97fc200edb74ce602960af0eaa354 Author: Hongyu Wang <[email protected]> Date: Tue Jan 27 15:38:12 2026 +0800 i386: Drop mask subst for define_insn_and_split of extend [PR123779] For define_insn_and_split, the subst applied only for the define_insn part, not the define_split part. So several define_insn_and_split with mask_name is actually producing non-splitable insns, resulting ICE in lra. Separate them to define_insn_and_split for mask/nonmask variants to generate corresponding splitters. PR target/123779 gcc/ChangeLog: * config/i386/sse.md (*sse4_1_<code>v8qiv8hi2<mask_name>_2): Rename to ... (*sse4_1_<code>v8qiv8hi2_2): ... this, and drop mask conditions. (*avx2_<code>v8qiv8si2<mask_name>_2): Rename to ... (*avx2_<code>v8qiv8si2_2): ... this, and likewise. (*sse4_1_<code>v4qiv4si2<mask_name>_2): Rename to ... (*sse4_1_<code>v4qiv4si2_2): ... this, and likewise. (*sse4_1_<code>v4hiv4si2<mask_name>_2): Rename to ... (*sse4_1_<code>v4hiv4si2_2): ... this, and likewise. (*avx2_<code>v4qiv4di2<mask_name>_2): Rename to ... (*avx2_<code>v4qiv4di2_2): ... this, and likewise. (*avx2_<code>v4hiv4di2<mask_name>_2): Rename to ... (*avx2_<code>v4hiv4di2_2): ... this, and likewise. (*sse4_1_<code>v2hiv2di2<mask_name>_2): Rename to ... (*sse4_1_<code>v2hiv2di2_2): ... this, and likewise. (*sse4_1_<code>v2siv2di2<mask_name>_2): Rename to ... (*sse4_1_<code>v2siv2di2_2): ... this, and likewise. (*avx512f_<code>v8qiv8di2<mask_name>_2): Rename to ... (*avx512f_<code>v8qiv8di2_2): ... this. (*avx512vl_<code>v8qiv8hi2_mask_2): New define_insn_and_split. (*avx512vl_<code>v8qiv8si2_mask_2): Likewise. (*avx512vl_<code>v4qiv4si2_mask_2): Likewise. (*avx512vl_<code>v4hiv4si2_mask_2): Likewise. (*avx512f_<code>v8qiv8di2_mask_2): Likewise. (*avx512vl_<code>v4qiv4di2_mask_2): Likewise. (*avx512vl_<code>v4hiv4di2_mask_2): Likewise. (*avx512vl_<code>v2hiv2di2_mask_2): Likewise. (*avx512vl_<code>v2siv2di2_mask_2): Likewise. gcc/testsuite/ChangeLog: * g++.target/i386/pr123779.C: New test. Diff: --- gcc/config/i386/sse.md | 270 +++++++++++++++++++++++++++++-- gcc/testsuite/g++.target/i386/pr123779.C | 56 +++++++ 2 files changed, 310 insertions(+), 16 deletions(-) diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md index 4bfbd3c59a99..cfe7a046f428 100644 --- a/gcc/config/i386/sse.md +++ b/gcc/config/i386/sse.md @@ -24599,7 +24599,7 @@ (set_attr "prefix" "orig,orig,maybe_evex") (set_attr "mode" "TI")]) -(define_insn_and_split "*sse4_1_<code>v8qiv8hi2<mask_name>_2" +(define_insn_and_split "*sse4_1_<code>v8qiv8hi2_2" [(set (match_operand:V8HI 0 "register_operand") (any_extend:V8HI (vec_select:V8QI @@ -24611,7 +24611,7 @@ (const_int 2) (const_int 3) (const_int 4) (const_int 5) (const_int 6) (const_int 7)]))))] - "TARGET_SSE4_1 && <mask_avx512bw_condition> && <mask_avx512vl_condition> + "TARGET_SSE4_1 && ix86_pre_reload_split ()" "#" "&& 1" @@ -24619,6 +24619,32 @@ (any_extend:V8HI (match_dup 1)))] "operands[1] = adjust_address_nv (operands[1], V8QImode, 0);") +(define_insn_and_split "*avx512vl_<code>v8qiv8hi2_mask_2" + [(set (match_operand:V8HI 0 "register_operand") + (vec_merge:V8HI + (any_extend:V8HI + (vec_select:V8QI + (subreg:V16QI + (vec_concat:V2DI + (match_operand:DI 1 "memory_operand") + (const_int 0)) 0) + (parallel [(const_int 0) (const_int 1) + (const_int 2) (const_int 3) + (const_int 4) (const_int 5) + (const_int 6) (const_int 7)]))) + (match_operand:V8HI 2 "nonimm_or_0_operand") + (match_operand:QI 3 "register_operand")))] + "TARGET_AVX512VL && TARGET_AVX512BW + && ix86_pre_reload_split ()" + "#" + "&& 1" + [(set (match_dup 0) + (vec_merge:V8HI + (any_extend:V8HI (match_dup 1)) + (match_dup 2) + (match_dup 3)))] + "operands[1] = adjust_address_nv (operands[1], V8QImode, 0);") + (define_insn_and_split "*sse4_1_zero_extendv8qiv8hi2_3" [(set (match_operand:V16QI 0 "register_operand" "=Yr,*x,Yw") (vec_select:V16QI @@ -24751,7 +24777,7 @@ (set_attr "prefix" "maybe_evex") (set_attr "mode" "OI")]) -(define_insn_and_split "*avx2_<code>v8qiv8si2<mask_name>_2" +(define_insn_and_split "*avx2_<code>v8qiv8si2_2" [(set (match_operand:V8SI 0 "register_operand") (any_extend:V8SI (vec_select:V8QI @@ -24763,7 +24789,7 @@ (const_int 2) (const_int 3) (const_int 4) (const_int 5) (const_int 6) (const_int 7)]))))] - "TARGET_AVX2 && <mask_avx512vl_condition> + "TARGET_AVX2 && ix86_pre_reload_split ()" "#" "&& 1" @@ -24771,6 +24797,33 @@ (any_extend:V8SI (match_dup 1)))] "operands[1] = adjust_address_nv (operands[1], V8QImode, 0);") +(define_insn_and_split "*avx512vl_<code>v8qiv8si2_mask_2" + [(set (match_operand:V8SI 0 "register_operand") + (vec_merge:V8SI + (any_extend:V8SI + (vec_select:V8QI + (subreg:V16QI + (vec_concat:V2DI + (match_operand:DI 1 "memory_operand") + (const_int 0)) 0) + (parallel [(const_int 0) (const_int 1) + (const_int 2) (const_int 3) + (const_int 4) (const_int 5) + (const_int 6) (const_int 7)]))) + (match_operand:V8SI 2 "nonimm_or_0_operand") + (match_operand:QI 3 "register_operand")))] + "TARGET_AVX512VL + && ix86_pre_reload_split ()" + "#" + "&& 1" + [(set (match_dup 0) + (vec_merge:V8SI + (any_extend:V8SI (match_dup 1)) + (match_dup 2) + (match_dup 3)))] + "operands[1] = adjust_address_nv (operands[1], V8QImode, 0);") + + (define_expand "<insn>v8qiv8si2" [(set (match_operand:V8SI 0 "register_operand") (any_extend:V8SI @@ -24814,7 +24867,7 @@ (set_attr "prefix" "orig,orig,maybe_evex") (set_attr "mode" "TI")]) -(define_insn_and_split "*sse4_1_<code>v4qiv4si2<mask_name>_2" +(define_insn_and_split "*sse4_1_<code>v4qiv4si2_2" [(set (match_operand:V4SI 0 "register_operand") (any_extend:V4SI (vec_select:V4QI @@ -24836,6 +24889,35 @@ (any_extend:V4SI (match_dup 1)))] "operands[1] = adjust_address_nv (operands[1], V4QImode, 0);") +(define_insn_and_split "*avx512vl_<code>v4qiv4si2_mask_2" + [(set (match_operand:V4SI 0 "register_operand") + (vec_merge:V4SI + (any_extend:V4SI + (vec_select:V4QI + (subreg:V16QI + (vec_merge:V4SI + (vec_duplicate:V4SI + (match_operand:SI 1 "memory_operand")) + (const_vector:V4SI + [(const_int 0) (const_int 0) + (const_int 0) (const_int 0)]) + (const_int 1)) 0) + (parallel [(const_int 0) (const_int 1) + (const_int 2) (const_int 3)]))) + (match_operand:V4SI 2 "nonimm_or_0_operand") + (match_operand:QI 3 "register_operand")))] + "TARGET_AVX512VL + && ix86_pre_reload_split ()" + "#" + "&& 1" + [(set (match_dup 0) + (vec_merge:V4SI + (any_extend:V4SI (match_dup 1)) + (match_dup 2) + (match_dup 3)))] + "operands[1] = adjust_address_nv (operands[1], V4QImode, 0);") + + (define_expand "<insn>v4qiv4si2" [(set (match_operand:V4SI 0 "register_operand") (any_extend:V4SI @@ -24987,7 +25069,7 @@ (set_attr "prefix" "orig,orig,maybe_evex") (set_attr "mode" "TI")]) -(define_insn_and_split "*sse4_1_<code>v4hiv4si2<mask_name>_2" +(define_insn_and_split "*sse4_1_<code>v4hiv4si2_2" [(set (match_operand:V4SI 0 "register_operand") (any_extend:V4SI (vec_select:V4HI @@ -24997,7 +25079,7 @@ (const_int 0)) 0) (parallel [(const_int 0) (const_int 1) (const_int 2) (const_int 3)]))))] - "TARGET_SSE4_1 && <mask_avx512vl_condition> + "TARGET_SSE4_1 && ix86_pre_reload_split ()" "#" "&& 1" @@ -25005,6 +25087,31 @@ (any_extend:V4SI (match_dup 1)))] "operands[1] = adjust_address_nv (operands[1], V4HImode, 0);") +(define_insn_and_split "*avx512vl_<code>v4hiv4si2_mask_2" + [(set (match_operand:V4SI 0 "register_operand") + (vec_merge:V4SI + (any_extend:V4SI + (vec_select:V4HI + (subreg:V8HI + (vec_concat:V2DI + (match_operand:DI 1 "memory_operand") + (const_int 0)) 0) + (parallel [(const_int 0) (const_int 1) + (const_int 2) (const_int 3)]))) + (match_operand:V4SI 2 "nonimm_or_0_operand") + (match_operand:QI 3 "register_operand")))] + "TARGET_AVX512VL + && ix86_pre_reload_split ()" + "#" + "&& 1" + [(set (match_dup 0) + (vec_merge:V4SI + (any_extend:V4SI (match_dup 1)) + (match_dup 2) + (match_dup 3)))] + "operands[1] = adjust_address_nv (operands[1], V4HImode, 0);") + + (define_expand "<insn>v4hiv4si2" [(set (match_operand:V4SI 0 "register_operand") (any_extend:V4SI @@ -25115,7 +25222,7 @@ (set_attr "prefix" "evex") (set_attr "mode" "XI")]) -(define_insn_and_split "*avx512f_<code>v8qiv8di2<mask_name>_2" +(define_insn_and_split "*avx512f_<code>v8qiv8di2_2" [(set (match_operand:V8DI 0 "register_operand") (any_extend:V8DI (vec_select:V8QI @@ -25134,6 +25241,32 @@ (any_extend:V8DI (match_dup 1)))] "operands[1] = adjust_address_nv (operands[1], V8QImode, 0);") +(define_insn_and_split "*avx512f_<code>v8qiv8di2_mask_2" + [(set (match_operand:V8DI 0 "register_operand") + (vec_merge:V8DI + (any_extend:V8DI + (vec_select:V8QI + (subreg:V16QI + (vec_concat:V2DI + (match_operand:DI 1 "memory_operand") + (const_int 0)) 0) + (parallel [(const_int 0) (const_int 1) + (const_int 2) (const_int 3) + (const_int 4) (const_int 5) + (const_int 6) (const_int 7)]))) + (match_operand:V8DI 2 "nonimm_or_0_operand") + (match_operand:QI 3 "register_operand")))] + "TARGET_AVX512F && ix86_pre_reload_split ()" + "#" + "&& 1" + [(set (match_dup 0) + (vec_merge:V8DI + (any_extend:V8DI (match_dup 1)) + (match_dup 2) + (match_dup 3)))] + "operands[1] = adjust_address_nv (operands[1], V8QImode, 0);") + + (define_expand "<insn>v8qiv8di2" [(set (match_operand:V8DI 0 "register_operand") (any_extend:V8DI @@ -25174,7 +25307,7 @@ (set_attr "prefix" "maybe_evex") (set_attr "mode" "OI")]) -(define_insn_and_split "*avx2_<code>v4qiv4di2<mask_name>_2" +(define_insn_and_split "*avx2_<code>v4qiv4di2_2" [(set (match_operand:V4DI 0 "register_operand") (any_extend:V4DI (vec_select:V4QI @@ -25188,7 +25321,7 @@ (const_int 1)) 0) (parallel [(const_int 0) (const_int 1) (const_int 2) (const_int 3)]))))] - "TARGET_AVX2 && <mask_avx512vl_condition> + "TARGET_AVX2 && ix86_pre_reload_split ()" "#" "&& 1" @@ -25196,6 +25329,35 @@ (any_extend:V4DI (match_dup 1)))] "operands[1] = adjust_address_nv (operands[1], V4QImode, 0);") +(define_insn_and_split "*avx512vl_<code>v4qiv4di2_mask_2" + [(set (match_operand:V4DI 0 "register_operand") + (vec_merge:V4DI + (any_extend:V4DI + (vec_select:V4QI + (subreg:V16QI + (vec_merge:V4SI + (vec_duplicate:V4SI + (match_operand:SI 1 "memory_operand")) + (const_vector:V4SI + [(const_int 0) (const_int 0) + (const_int 0) (const_int 0)]) + (const_int 1)) 0) + (parallel [(const_int 0) (const_int 1) + (const_int 2) (const_int 3)]))) + (match_operand:V4DI 2 "nonimm_or_0_operand") + (match_operand:QI 3 "register_operand")))] + "TARGET_AVX512VL + && ix86_pre_reload_split ()" + "#" + "&& 1" + [(set (match_dup 0) + (vec_merge:V4DI + (any_extend:V4DI (match_dup 1)) + (match_dup 2) + (match_dup 3)))] + "operands[1] = adjust_address_nv (operands[1], V4QImode, 0);") + + (define_expand "<insn>v4qiv4di2" [(set (match_operand:V4DI 0 "register_operand") (any_extend:V4DI @@ -25317,7 +25479,7 @@ (set_attr "prefix" "maybe_evex") (set_attr "mode" "OI")]) -(define_insn_and_split "*avx2_<code>v4hiv4di2<mask_name>_2" +(define_insn_and_split "*avx2_<code>v4hiv4di2_2" [(set (match_operand:V4DI 0 "register_operand") (any_extend:V4DI (vec_select:V4HI @@ -25327,7 +25489,7 @@ (const_int 0)) 0) (parallel [(const_int 0) (const_int 1) (const_int 2) (const_int 3)]))))] - "TARGET_AVX2 && <mask_avx512vl_condition> + "TARGET_AVX2 && ix86_pre_reload_split ()" "#" "&& 1" @@ -25335,6 +25497,31 @@ (any_extend:V4DI (match_dup 1)))] "operands[1] = adjust_address_nv (operands[1], V4HImode, 0);") +(define_insn_and_split "*avx512vl_<code>v4hiv4di2_mask_2" + [(set (match_operand:V4DI 0 "register_operand") + (vec_merge:V4DI + (any_extend:V4DI + (vec_select:V4HI + (subreg:V8HI + (vec_concat:V2DI + (match_operand:DI 1 "memory_operand") + (const_int 0)) 0) + (parallel [(const_int 0) (const_int 1) + (const_int 2) (const_int 3)]))) + (match_operand:V4DI 2 "nonimm_or_0_operand") + (match_operand:QI 3 "register_operand")))] + "TARGET_AVX512VL + && ix86_pre_reload_split ()" + "#" + "&& 1" + [(set (match_dup 0) + (vec_merge:V4DI + (any_extend:V4DI (match_dup 1)) + (match_dup 2) + (match_dup 3)))] + "operands[1] = adjust_address_nv (operands[1], V4HImode, 0);") + + (define_expand "<insn>v4hiv4di2" [(set (match_operand:V4DI 0 "register_operand") (any_extend:V4DI @@ -25377,7 +25564,7 @@ (set_attr "prefix" "orig,orig,maybe_evex") (set_attr "mode" "TI")]) -(define_insn_and_split "*sse4_1_<code>v2hiv2di2<mask_name>_2" +(define_insn_and_split "*sse4_1_<code>v2hiv2di2_2" [(set (match_operand:V2DI 0 "register_operand") (any_extend:V2DI (vec_select:V2HI @@ -25390,7 +25577,7 @@ (const_int 0) (const_int 0)]) (const_int 1)) 0) (parallel [(const_int 0) (const_int 1)]))))] - "TARGET_SSE4_1 && <mask_avx512vl_condition> + "TARGET_SSE4_1 && ix86_pre_reload_split ()" "#" "&& 1" @@ -25398,6 +25585,34 @@ (any_extend:V2DI (match_dup 1)))] "operands[1] = adjust_address_nv (operands[1], V2HImode, 0);") +(define_insn_and_split "*avx512vl_<code>v2hiv2di2_mask_2" + [(set (match_operand:V2DI 0 "register_operand") + (vec_merge:V2DI + (any_extend:V2DI + (vec_select:V2HI + (subreg:V8HI + (vec_merge:V4SI + (vec_duplicate:V4SI + (match_operand:SI 1 "memory_operand")) + (const_vector:V4SI + [(const_int 0) (const_int 0) + (const_int 0) (const_int 0)]) + (const_int 1)) 0) + (parallel [(const_int 0) (const_int 1)]))) + (match_operand:V2DI 2 "nonimm_or_0_operand") + (match_operand:QI 3 "register_operand")))] + "TARGET_AVX512VL + && ix86_pre_reload_split ()" + "#" + "&& 1" + [(set (match_dup 0) + (vec_merge:V2DI + (any_extend:V2DI (match_dup 1)) + (match_dup 2) + (match_dup 3)))] + "operands[1] = adjust_address_nv (operands[1], V2HImode, 0);") + + (define_expand "<insn>v2hiv2di2" [(set (match_operand:V2DI 0 "register_operand") (any_extend:V2DI @@ -25543,7 +25758,7 @@ (set_attr "prefix" "orig,orig,maybe_evex") (set_attr "mode" "TI")]) -(define_insn_and_split "*sse4_1_<code>v2siv2di2<mask_name>_2" +(define_insn_and_split "*sse4_1_<code>v2siv2di2_2" [(set (match_operand:V2DI 0 "register_operand") (any_extend:V2DI (vec_select:V2SI @@ -25552,7 +25767,7 @@ (match_operand:DI 1 "memory_operand") (const_int 0)) 0) (parallel [(const_int 0) (const_int 1)]))))] - "TARGET_SSE4_1 && <mask_avx512vl_condition> + "TARGET_SSE4_1 && ix86_pre_reload_split ()" "#" "&& 1" @@ -25560,6 +25775,29 @@ (any_extend:V2DI (match_dup 1)))] "operands[1] = adjust_address_nv (operands[1], V2SImode, 0);") +(define_insn_and_split "*avx512vl_<code>v2siv2di2_mask_2" + [(set (match_operand:V2DI 0 "register_operand") + (vec_merge:V2DI + (any_extend:V2DI + (vec_select:V2SI + (subreg:V4SI + (vec_concat:V2DI + (match_operand:DI 1 "memory_operand") + (const_int 0)) 0) + (parallel [(const_int 0) (const_int 1)]))) + (match_operand:V2DI 2 "nonimm_or_0_operand") + (match_operand:QI 3 "register_operand")))] + "TARGET_AVX512VL + && ix86_pre_reload_split ()" + "#" + "&& 1" + [(set (match_dup 0) + (vec_merge:V2DI + (any_extend:V2DI (match_dup 1)) + (match_dup 2) + (match_dup 3)))] + "operands[1] = adjust_address_nv (operands[1], V2SImode, 0);") + (define_insn_and_split "*sse4_1_zero_extendv2siv2di2_3" [(set (match_operand:V4SI 0 "register_operand" "=Yr,*x,v") (vec_select:V4SI diff --git a/gcc/testsuite/g++.target/i386/pr123779.C b/gcc/testsuite/g++.target/i386/pr123779.C new file mode 100644 index 000000000000..e0b904481090 --- /dev/null +++ b/gcc/testsuite/g++.target/i386/pr123779.C @@ -0,0 +1,56 @@ +/* PR target/123779 */ +/* { dg-do compile } */ +/* { dg-options "-O2 -mavx512vl -mavx512bw" } */ + +typedef char __v16qi __attribute__((__vector_size__(16))); +typedef short __v8hi __attribute__((__vector_size__(16))); +typedef int __v4si __attribute__((__vector_size__(16))); +typedef long long __v2di __attribute__((__vector_size__(16))); + +typedef int __v8si __attribute__((__vector_size__(32))); +typedef long long __v4di __attribute__((__vector_size__(32))); +typedef long long __v8di __attribute__((__vector_size__(64))); +typedef unsigned char __mmask8; + +long long g_mem; + +#define MAKE_TEST_BLEND(NAME, DEST_T, SRC_INIT_T, PMOV_INPUT_CAST_T, \ + PMOV_FUNC, BLEND_FUNC, BLEND_CAST_T, LOAD_VAL_TYPE, ...) \ + DEST_T dest_##NAME, mask_src1_##NAME, mask_src2_##NAME, mask_##NAME, res_##NAME; \ + void test_##NAME() \ + { \ + mask_##NAME = mask_src1_##NAME < mask_src2_##NAME; \ + LOAD_VAL_TYPE val = (LOAD_VAL_TYPE)g_mem; \ + SRC_INIT_T src_vec = { __VA_ARGS__ }; \ + DEST_T extended = (DEST_T)PMOV_FUNC ((PMOV_INPUT_CAST_T)src_vec); \ + res_##NAME = (DEST_T)BLEND_FUNC ( \ + (BLEND_CAST_T)dest_##NAME, \ + (BLEND_CAST_T)extended, \ + (BLEND_CAST_T)mask_##NAME); \ + } + +#define MAKE_TEST_MASK(NAME, DEST_T, SRC_INIT_T, PMOV_INPUT_CAST_T, \ + PMOV_MASK_FUNC, LOAD_VAL_TYPE, ...) \ + DEST_T dest_##NAME, res_##NAME; \ + __mmask8 mask_##NAME; \ + void test_##NAME() \ + { \ + LOAD_VAL_TYPE val = (LOAD_VAL_TYPE) g_mem; \ + SRC_INIT_T src_vec = { __VA_ARGS__ }; \ + res_##NAME = PMOV_MASK_FUNC ((PMOV_INPUT_CAST_T)src_vec, \ + dest_##NAME, \ + mask_##NAME);\ + } + + +MAKE_TEST_BLEND(v8qi_v8hi, __v8hi, __v2di, __v16qi, __builtin_ia32_pmovzxbw128, __builtin_ia32_pblendvb128, __v16qi, long long, val, 0) +MAKE_TEST_BLEND(v4qi_v4si, __v4si, __v4si, __v16qi, __builtin_ia32_pmovzxbd128, __builtin_ia32_pblendvb128, __v16qi, int, val, 0, 0, 0) +MAKE_TEST_BLEND(v4hi_v4si, __v4si, __v2di, __v8hi, __builtin_ia32_pmovzxwd128, __builtin_ia32_pblendvb128, __v16qi, long long, val, 0) +MAKE_TEST_BLEND(v2hi_v2di, __v2di, __v4si, __v8hi, __builtin_ia32_pmovzxwq128, __builtin_ia32_pblendvb128, __v16qi, int, val, 0, 0, 0) +MAKE_TEST_BLEND(v2si_v2di, __v2di, __v2di, __v4si, __builtin_ia32_pmovzxdq128, __builtin_ia32_pblendvb128, __v16qi, long long, val, 0) + + +MAKE_TEST_MASK(v8qi_v8si, __v8si, __v2di, __v16qi, __builtin_ia32_pmovzxbd256_mask, long long, val, 0) +MAKE_TEST_MASK(v4qi_v4di, __v4di, __v4si, __v16qi, __builtin_ia32_pmovzxbq256_mask, int, val, 0, 0, 0) +MAKE_TEST_MASK(v4hi_v4di, __v4di, __v2di, __v8hi, __builtin_ia32_pmovzxwq256_mask, long long, val, 0) +MAKE_TEST_MASK(v8qi_v8di, __v8di, __v2di, __v16qi, __builtin_ia32_pmovzxbq512_mask, long long, val, 0)
