@@ -1952,7 +1952,7 @@ define void @vec256_i32_widen_to_i64_factor2_broadcast_to_v4i64_factor4(ptr %in.
19521952; AVX2-NEXT:    vpaddb 32(%rsi), %ymm0, %ymm0 
19531953; AVX2-NEXT:    vmovdqa (%rdi), %xmm1 
19541954; AVX2-NEXT:    vpaddb (%rsi), %xmm1, %xmm1 
1955- ; AVX2-NEXT:    vpbroadcastq  %xmm1, %ymm1 
1955+ ; AVX2-NEXT:    vpbroadcastd  %xmm1, %ymm1 
19561956; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2],ymm0[3],ymm1[4],ymm0[5],ymm1[6],ymm0[7] 
19571957; AVX2-NEXT:    vpaddb (%rdx), %ymm0, %ymm0 
19581958; AVX2-NEXT:    vmovdqa %ymm0, (%rcx) 
@@ -1965,7 +1965,7 @@ define void @vec256_i32_widen_to_i64_factor2_broadcast_to_v4i64_factor4(ptr %in.
19651965; AVX512F-SLOW-NEXT:    vpaddb 32(%rsi), %ymm0, %ymm0 
19661966; AVX512F-SLOW-NEXT:    vmovdqa (%rdi), %xmm1 
19671967; AVX512F-SLOW-NEXT:    vpaddb (%rsi), %xmm1, %xmm1 
1968- ; AVX512F-SLOW-NEXT:    vpbroadcastq  %xmm1, %ymm1 
1968+ ; AVX512F-SLOW-NEXT:    vpbroadcastd  %xmm1, %ymm1 
19691969; AVX512F-SLOW-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2],ymm0[3],ymm1[4],ymm0[5],ymm1[6],ymm0[7] 
19701970; AVX512F-SLOW-NEXT:    vpaddb (%rdx), %ymm0, %ymm0 
19711971; AVX512F-SLOW-NEXT:    vmovdqa %ymm0, (%rcx) 
@@ -1991,7 +1991,7 @@ define void @vec256_i32_widen_to_i64_factor2_broadcast_to_v4i64_factor4(ptr %in.
19911991; AVX512DQ-SLOW-NEXT:    vpaddb 32(%rsi), %ymm0, %ymm0 
19921992; AVX512DQ-SLOW-NEXT:    vmovdqa (%rdi), %xmm1 
19931993; AVX512DQ-SLOW-NEXT:    vpaddb (%rsi), %xmm1, %xmm1 
1994- ; AVX512DQ-SLOW-NEXT:    vpbroadcastq  %xmm1, %ymm1 
1994+ ; AVX512DQ-SLOW-NEXT:    vpbroadcastd  %xmm1, %ymm1 
19951995; AVX512DQ-SLOW-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2],ymm0[3],ymm1[4],ymm0[5],ymm1[6],ymm0[7] 
19961996; AVX512DQ-SLOW-NEXT:    vpaddb (%rdx), %ymm0, %ymm0 
19971997; AVX512DQ-SLOW-NEXT:    vmovdqa %ymm0, (%rcx) 
@@ -2016,7 +2016,7 @@ define void @vec256_i32_widen_to_i64_factor2_broadcast_to_v4i64_factor4(ptr %in.
20162016; AVX512BW-SLOW-NEXT:    vmovdqa64 (%rdi), %zmm0 
20172017; AVX512BW-SLOW-NEXT:    vpaddb (%rsi), %zmm0, %zmm0 
20182018; AVX512BW-SLOW-NEXT:    vextracti64x4 $1, %zmm0, %ymm1 
2019- ; AVX512BW-SLOW-NEXT:    vpbroadcastq  %xmm0, %ymm0 
2019+ ; AVX512BW-SLOW-NEXT:    vpbroadcastd  %xmm0, %ymm0 
20202020; AVX512BW-SLOW-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7] 
20212021; AVX512BW-SLOW-NEXT:    vpaddb (%rdx), %zmm0, %zmm0 
20222022; AVX512BW-SLOW-NEXT:    vmovdqa64 %zmm0, (%rcx) 
@@ -4649,11 +4649,10 @@ define void @vec384_i32_widen_to_i64_factor2_broadcast_to_v6i64_factor6(ptr %in.
46494649; AVX2-SLOW-NEXT:    vmovdqa 48(%rdi), %xmm1 
46504650; AVX2-SLOW-NEXT:    vpaddb 48(%rsi), %xmm1, %xmm1 
46514651; AVX2-SLOW-NEXT:    vpaddb (%rsi), %xmm0, %xmm0 
4652- ; AVX2-SLOW-NEXT:    vpbroadcastq  %xmm0, %ymm2  
4653- ; AVX2-SLOW-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2 [0],ymm1[1],ymm2 [2],ymm1[3],ymm2 [4,5,6,7] 
4652+ ; AVX2-SLOW-NEXT:    vpbroadcastd  %xmm0, %ymm0  
4653+ ; AVX2-SLOW-NEXT:    vpblendd {{.*#+}} ymm1 = ymm0 [0],ymm1[1],ymm0 [2],ymm1[3],ymm0 [4,5,6,7] 
46544654; AVX2-SLOW-NEXT:    vpxor %xmm2, %xmm2, %xmm2 
46554655; AVX2-SLOW-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5],ymm1[6],ymm2[7] 
4656- ; AVX2-SLOW-NEXT:    vpbroadcastd %xmm0, %xmm0 
46574656; AVX2-SLOW-NEXT:    vpxor %xmm2, %xmm2, %xmm2 
46584657; AVX2-SLOW-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3] 
46594658; AVX2-SLOW-NEXT:    vpaddb (%rdx), %ymm1, %ymm1 
@@ -4669,7 +4668,7 @@ define void @vec384_i32_widen_to_i64_factor2_broadcast_to_v6i64_factor6(ptr %in.
46694668; AVX2-FAST-PERLANE-NEXT:    vmovdqa 48(%rdi), %xmm1 
46704669; AVX2-FAST-PERLANE-NEXT:    vpaddb 48(%rsi), %xmm1, %xmm1 
46714670; AVX2-FAST-PERLANE-NEXT:    vpaddb (%rsi), %xmm0, %xmm0 
4672- ; AVX2-FAST-PERLANE-NEXT:    vpbroadcastq  %xmm0, %ymm2 
4671+ ; AVX2-FAST-PERLANE-NEXT:    vpbroadcastd  %xmm0, %ymm2 
46734672; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2],ymm1[3],ymm2[4,5,6,7] 
46744673; AVX2-FAST-PERLANE-NEXT:    vpxor %xmm2, %xmm2, %xmm2 
46754674; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5],ymm1[6],ymm2[7] 
@@ -4687,7 +4686,7 @@ define void @vec384_i32_widen_to_i64_factor2_broadcast_to_v6i64_factor6(ptr %in.
46874686; AVX2-FAST-NEXT:    vmovdqa 48(%rdi), %xmm1 
46884687; AVX2-FAST-NEXT:    vpaddb 48(%rsi), %xmm1, %xmm1 
46894688; AVX2-FAST-NEXT:    vpaddb (%rsi), %xmm0, %xmm0 
4690- ; AVX2-FAST-NEXT:    vpbroadcastq  %xmm0, %ymm2 
4689+ ; AVX2-FAST-NEXT:    vpbroadcastd  %xmm0, %ymm2 
46914690; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2],ymm1[3],ymm2[4,5,6,7] 
46924691; AVX2-FAST-NEXT:    vpxor %xmm2, %xmm2, %xmm2 
46934692; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5],ymm1[6],ymm2[7] 
@@ -6582,7 +6581,7 @@ define void @vec512_i32_widen_to_i64_factor2_broadcast_to_v8i64_factor8(ptr %in.
65826581; AVX2-NEXT:    vpxor %xmm0, %xmm0, %xmm0 
65836582; AVX2-NEXT:    vmovdqa (%rdi), %xmm1 
65846583; AVX2-NEXT:    vpaddb (%rsi), %xmm1, %xmm1 
6585- ; AVX2-NEXT:    vpbroadcastq  %xmm1, %ymm1 
6584+ ; AVX2-NEXT:    vpbroadcastd  %xmm1, %ymm1 
65866585; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2],ymm0[3],ymm1[4],ymm0[5],ymm1[6],ymm0[7] 
65876586; AVX2-NEXT:    vpaddb 32(%rdx), %ymm0, %ymm1 
65886587; AVX2-NEXT:    vpaddb (%rdx), %ymm0, %ymm0 
0 commit comments