Page MenuHomePhabricator
Differential D79096 Diff 261671 llvm/test/CodeGen/X86/bitcast-vector-bool.ll

Changeset View

Standalone View

View Options

llvm/test/CodeGen/X86/bitcast-vector-bool.ll

Show First 20 Lines • Show All 43 Lines • ▼ Show 20 Lines; AVX512-NEXT: retq
ret i1 %5 ret i1 %5
} }
define i2 @bitcast_v4i32_to_v2i2(<4 x i32> %a0) nounwind { define i2 @bitcast_v4i32_to_v2i2(<4 x i32> %a0) nounwind {
; SSE2-SSSE3-LABEL: bitcast_v4i32_to_v2i2: ; SSE2-SSSE3-LABEL: bitcast_v4i32_to_v2i2:
; SSE2-SSSE3: # %bb.0: ; SSE2-SSSE3: # %bb.0:
; SSE2-SSSE3-NEXT: movmskps %xmm0, %eax ; SSE2-SSSE3-NEXT: movmskps %xmm0, %eax
; SSE2-SSSE3-NEXT: movl %eax, %ecx ; SSE2-SSSE3-NEXT: movl %eax, %ecx
; SSE2-SSSE3-NEXT: andl $3, %ecx ; SSE2-SSSE3-NEXT: shrb $2, %cl
; SSE2-SSSE3-NEXT: movq %rcx, %xmm0 ; SSE2-SSSE3-NEXT: movzbl %cl, %ecx
; SSE2-SSSE3-NEXT: shrl $2, %eax ; SSE2-SSSE3-NEXT: andb $3, %al
; SSE2-SSSE3-NEXT: movq %rax, %xmm1 ; SSE2-SSSE3-NEXT: movzbl %al, %eax
; SSE2-SSSE3-NEXT: punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0] ; SSE2-SSSE3-NEXT: movd %eax, %xmm0
; SSE2-SSSE3-NEXT: pinsrw $4, %ecx, %xmm0
; SSE2-SSSE3-NEXT: movdqa %xmm0, -{{[0-9]+}}(%rsp) ; SSE2-SSSE3-NEXT: movdqa %xmm0, -{{[0-9]+}}(%rsp)
; SSE2-SSSE3-NEXT: movb -{{[0-9]+}}(%rsp), %al ; SSE2-SSSE3-NEXT: movb -{{[0-9]+}}(%rsp), %al
; SSE2-SSSE3-NEXT: addb -{{[0-9]+}}(%rsp), %al ; SSE2-SSSE3-NEXT: addb -{{[0-9]+}}(%rsp), %al
; SSE2-SSSE3-NEXT: retq ; SSE2-SSSE3-NEXT: retq
; ;
; AVX12-LABEL: bitcast_v4i32_to_v2i2: ; AVX12-LABEL: bitcast_v4i32_to_v2i2:
; AVX12: # %bb.0: ; AVX12: # %bb.0:
; AVX12-NEXT: vmovmskps %xmm0, %ecx ; AVX12-NEXT: vmovmskps %xmm0, %eax
; AVX12-NEXT: movl %ecx, %eax ; AVX12-NEXT: movl %eax, %ecx
; AVX12-NEXT: shrl $2, %eax ; AVX12-NEXT: shrb $2, %cl
; AVX12-NEXT: andl $3, %ecx ; AVX12-NEXT: andb $3, %al
; AVX12-NEXT: addb %cl, %al ; AVX12-NEXT: addb %cl, %al
; AVX12-NEXT: # kill: def $al killed $al killed $eax ; AVX12-NEXT: # kill: def $al killed $al killed $eax
; AVX12-NEXT: retq ; AVX12-NEXT: retq
; ;
; AVX512-LABEL: bitcast_v4i32_to_v2i2: ; AVX512-LABEL: bitcast_v4i32_to_v2i2:
; AVX512: # %bb.0: ; AVX512: # %bb.0:
; AVX512-NEXT: vpxor %xmm1, %xmm1, %xmm1 ; AVX512-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512-NEXT: vpcmpgtd %xmm0, %xmm1, %k0 ; AVX512-NEXT: vpcmpgtd %xmm0, %xmm1, %k0
; AVX512-NEXT: kmovd %k0, %ecx ; AVX512-NEXT: kmovd %k0, %eax
; AVX512-NEXT: movzbl %cl, %eax ; AVX512-NEXT: movl %eax, %ecx
; AVX512-NEXT: shrl $2, %eax ; AVX512-NEXT: shrb $2, %cl
; AVX512-NEXT: andl $3, %eax ; AVX512-NEXT: andb $3, %al
; AVX512-NEXT: andl $3, %ecx
; AVX512-NEXT: addb %cl, %al ; AVX512-NEXT: addb %cl, %al
; AVX512-NEXT: # kill: def $al killed $al killed $eax ; AVX512-NEXT: # kill: def $al killed $al killed $eax
; AVX512-NEXT: retq ; AVX512-NEXT: retq
%1 = icmp slt <4 x i32> %a0, zeroinitializer %1 = icmp slt <4 x i32> %a0, zeroinitializer
%2 = bitcast <4 x i1> %1 to <2 x i2> %2 = bitcast <4 x i1> %1 to <2 x i2>
%3 = extractelement <2 x i2> %2, i32 0 %3 = extractelement <2 x i2> %2, i32 0
%4 = extractelement <2 x i2> %2, i32 1 %4 = extractelement <2 x i2> %2, i32 1
%5 = add i2 %3, %4 %5 = add i2 %3, %4
ret i2 %5 ret i2 %5
} }
define i4 @bitcast_v8i16_to_v2i4(<8 x i16> %a0) nounwind { define i4 @bitcast_v8i16_to_v2i4(<8 x i16> %a0) nounwind {
; SSE2-SSSE3-LABEL: bitcast_v8i16_to_v2i4: ; SSE2-SSSE3-LABEL: bitcast_v8i16_to_v2i4:
; SSE2-SSSE3: # %bb.0: ; SSE2-SSSE3: # %bb.0:
; SSE2-SSSE3-NEXT: packsswb %xmm0, %xmm0 ; SSE2-SSSE3-NEXT: packsswb %xmm0, %xmm0
; SSE2-SSSE3-NEXT: pmovmskb %xmm0, %eax ; SSE2-SSSE3-NEXT: pmovmskb %xmm0, %eax
; SSE2-SSSE3-NEXT: movzbl %al, %ecx ; SSE2-SSSE3-NEXT: movl %eax, %ecx
; SSE2-SSSE3-NEXT: shrl $4, %ecx ; SSE2-SSSE3-NEXT: shrb $4, %cl
; SSE2-SSSE3-NEXT: movq %rcx, %xmm0 ; SSE2-SSSE3-NEXT: movzbl %cl, %ecx
; SSE2-SSSE3-NEXT: andl $15, %eax ; SSE2-SSSE3-NEXT: andb $15, %al
; SSE2-SSSE3-NEXT: movq %rax, %xmm1 ; SSE2-SSSE3-NEXT: movzbl %al, %eax
; SSE2-SSSE3-NEXT: punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0] ; SSE2-SSSE3-NEXT: movd %eax, %xmm0
; SSE2-SSSE3-NEXT: movdqa %xmm1, -{{[0-9]+}}(%rsp) ; SSE2-SSSE3-NEXT: pinsrw $4, %ecx, %xmm0
; SSE2-SSSE3-NEXT: movdqa %xmm0, -{{[0-9]+}}(%rsp)
; SSE2-SSSE3-NEXT: movb -{{[0-9]+}}(%rsp), %al ; SSE2-SSSE3-NEXT: movb -{{[0-9]+}}(%rsp), %al
; SSE2-SSSE3-NEXT: addb -{{[0-9]+}}(%rsp), %al ; SSE2-SSSE3-NEXT: addb -{{[0-9]+}}(%rsp), %al
; SSE2-SSSE3-NEXT: retq ; SSE2-SSSE3-NEXT: retq
; ;
; AVX12-LABEL: bitcast_v8i16_to_v2i4: ; AVX12-LABEL: bitcast_v8i16_to_v2i4:
; AVX12: # %bb.0: ; AVX12: # %bb.0:
; AVX12-NEXT: vpacksswb %xmm0, %xmm0, %xmm0 ; AVX12-NEXT: vpacksswb %xmm0, %xmm0, %xmm0
; AVX12-NEXT: vpmovmskb %xmm0, %ecx ; AVX12-NEXT: vpmovmskb %xmm0, %eax
; AVX12-NEXT: movzbl %cl, %eax ; AVX12-NEXT: movl %eax, %ecx
; AVX12-NEXT: shrl $4, %eax ; AVX12-NEXT: shrb $4, %cl
; AVX12-NEXT: andl $15, %ecx ; AVX12-NEXT: andb $15, %al
; AVX12-NEXT: addb %cl, %al ; AVX12-NEXT: addb %cl, %al
; AVX12-NEXT: # kill: def $al killed $al killed $eax ; AVX12-NEXT: # kill: def $al killed $al killed $eax
; AVX12-NEXT: retq ; AVX12-NEXT: retq
; ;
; AVX512-LABEL: bitcast_v8i16_to_v2i4: ; AVX512-LABEL: bitcast_v8i16_to_v2i4:
; AVX512: # %bb.0: ; AVX512: # %bb.0:
; AVX512-NEXT: vpmovw2m %xmm0, %k0 ; AVX512-NEXT: vpmovw2m %xmm0, %k0
; AVX512-NEXT: kmovd %k0, %ecx ; AVX512-NEXT: kmovd %k0, %eax
; AVX512-NEXT: movzbl %cl, %eax ; AVX512-NEXT: movl %eax, %ecx
; AVX512-NEXT: shrl $4, %eax ; AVX512-NEXT: shrb $4, %cl
; AVX512-NEXT: andl $15, %ecx ; AVX512-NEXT: andb $15, %al
; AVX512-NEXT: addb %cl, %al ; AVX512-NEXT: addb %cl, %al
; AVX512-NEXT: # kill: def $al killed $al killed $eax ; AVX512-NEXT: # kill: def $al killed $al killed $eax
; AVX512-NEXT: retq ; AVX512-NEXT: retq
%1 = icmp slt <8 x i16> %a0, zeroinitializer %1 = icmp slt <8 x i16> %a0, zeroinitializer
%2 = bitcast <8 x i1> %1 to <2 x i4> %2 = bitcast <8 x i1> %1 to <2 x i4>
%3 = extractelement <2 x i4> %2, i32 0 %3 = extractelement <2 x i4> %2, i32 0
%4 = extractelement <2 x i4> %2, i32 1 %4 = extractelement <2 x i4> %2, i32 1
%5 = add i4 %3, %4 %5 = add i4 %3, %4
▲ Show 20 Lines • Show All 42 Lines • ▼ Show 20 Lines
; ;
define i2 @bitcast_v4i64_to_v2i2(<4 x i64> %a0) nounwind { define i2 @bitcast_v4i64_to_v2i2(<4 x i64> %a0) nounwind {
; SSE2-SSSE3-LABEL: bitcast_v4i64_to_v2i2: ; SSE2-SSSE3-LABEL: bitcast_v4i64_to_v2i2:
; SSE2-SSSE3: # %bb.0: ; SSE2-SSSE3: # %bb.0:
; SSE2-SSSE3-NEXT: packssdw %xmm1, %xmm0 ; SSE2-SSSE3-NEXT: packssdw %xmm1, %xmm0
; SSE2-SSSE3-NEXT: movmskps %xmm0, %eax ; SSE2-SSSE3-NEXT: movmskps %xmm0, %eax
; SSE2-SSSE3-NEXT: movl %eax, %ecx ; SSE2-SSSE3-NEXT: movl %eax, %ecx
; SSE2-SSSE3-NEXT: shrl $2, %ecx ; SSE2-SSSE3-NEXT: shrb $2, %cl
; SSE2-SSSE3-NEXT: movq %rcx, %xmm0 ; SSE2-SSSE3-NEXT: movzbl %cl, %ecx
; SSE2-SSSE3-NEXT: andl $3, %eax ; SSE2-SSSE3-NEXT: andb $3, %al
; SSE2-SSSE3-NEXT: movq %rax, %xmm1 ; SSE2-SSSE3-NEXT: movzbl %al, %eax
; SSE2-SSSE3-NEXT: punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0] ; SSE2-SSSE3-NEXT: movd %eax, %xmm0
; SSE2-SSSE3-NEXT: movdqa %xmm1, -{{[0-9]+}}(%rsp) ; SSE2-SSSE3-NEXT: pinsrw $4, %ecx, %xmm0
; SSE2-SSSE3-NEXT: movdqa %xmm0, -{{[0-9]+}}(%rsp)
; SSE2-SSSE3-NEXT: movb -{{[0-9]+}}(%rsp), %al ; SSE2-SSSE3-NEXT: movb -{{[0-9]+}}(%rsp), %al
; SSE2-SSSE3-NEXT: addb -{{[0-9]+}}(%rsp), %al ; SSE2-SSSE3-NEXT: addb -{{[0-9]+}}(%rsp), %al
; SSE2-SSSE3-NEXT: retq ; SSE2-SSSE3-NEXT: retq
; ;
; AVX12-LABEL: bitcast_v4i64_to_v2i2: ; AVX12-LABEL: bitcast_v4i64_to_v2i2:
; AVX12: # %bb.0: ; AVX12: # %bb.0:
; AVX12-NEXT: vmovmskpd %ymm0, %ecx ; AVX12-NEXT: vmovmskpd %ymm0, %eax
; AVX12-NEXT: movl %ecx, %eax ; AVX12-NEXT: movl %eax, %ecx
; AVX12-NEXT: shrl $2, %eax ; AVX12-NEXT: shrb $2, %cl
; AVX12-NEXT: andl $3, %ecx ; AVX12-NEXT: andb $3, %al
; AVX12-NEXT: addb %cl, %al ; AVX12-NEXT: addb %cl, %al
; AVX12-NEXT: # kill: def $al killed $al killed $eax ; AVX12-NEXT: # kill: def $al killed $al killed $eax
; AVX12-NEXT: vzeroupper ; AVX12-NEXT: vzeroupper
; AVX12-NEXT: retq ; AVX12-NEXT: retq
; ;
; AVX512-LABEL: bitcast_v4i64_to_v2i2: ; AVX512-LABEL: bitcast_v4i64_to_v2i2:
; AVX512: # %bb.0: ; AVX512: # %bb.0:
; AVX512-NEXT: vpxor %xmm1, %xmm1, %xmm1 ; AVX512-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512-NEXT: vpcmpgtq %ymm0, %ymm1, %k0 ; AVX512-NEXT: vpcmpgtq %ymm0, %ymm1, %k0
; AVX512-NEXT: kmovd %k0, %ecx ; AVX512-NEXT: kmovd %k0, %eax
; AVX512-NEXT: movzbl %cl, %eax ; AVX512-NEXT: movl %eax, %ecx
; AVX512-NEXT: shrl $2, %eax ; AVX512-NEXT: shrb $2, %cl
; AVX512-NEXT: andl $3, %eax ; AVX512-NEXT: andb $3, %al
; AVX512-NEXT: andl $3, %ecx
; AVX512-NEXT: addb %cl, %al ; AVX512-NEXT: addb %cl, %al
; AVX512-NEXT: # kill: def $al killed $al killed $eax ; AVX512-NEXT: # kill: def $al killed $al killed $eax
; AVX512-NEXT: vzeroupper ; AVX512-NEXT: vzeroupper
; AVX512-NEXT: retq ; AVX512-NEXT: retq
%1 = icmp slt <4 x i64> %a0, zeroinitializer %1 = icmp slt <4 x i64> %a0, zeroinitializer
%2 = bitcast <4 x i1> %1 to <2 x i2> %2 = bitcast <4 x i1> %1 to <2 x i2>
%3 = extractelement <2 x i2> %2, i32 0 %3 = extractelement <2 x i2> %2, i32 0
%4 = extractelement <2 x i2> %2, i32 1 %4 = extractelement <2 x i2> %2, i32 1
%5 = add i2 %3, %4 %5 = add i2 %3, %4
ret i2 %5 ret i2 %5
} }
define i4 @bitcast_v8i32_to_v2i4(<8 x i32> %a0) nounwind { define i4 @bitcast_v8i32_to_v2i4(<8 x i32> %a0) nounwind {
; SSE2-SSSE3-LABEL: bitcast_v8i32_to_v2i4: ; SSE2-SSSE3-LABEL: bitcast_v8i32_to_v2i4:
; SSE2-SSSE3: # %bb.0: ; SSE2-SSSE3: # %bb.0:
; SSE2-SSSE3-NEXT: packssdw %xmm1, %xmm0 ; SSE2-SSSE3-NEXT: packssdw %xmm1, %xmm0
; SSE2-SSSE3-NEXT: packsswb %xmm0, %xmm0 ; SSE2-SSSE3-NEXT: packsswb %xmm0, %xmm0
; SSE2-SSSE3-NEXT: pmovmskb %xmm0, %eax ; SSE2-SSSE3-NEXT: pmovmskb %xmm0, %eax
; SSE2-SSSE3-NEXT: movzbl %al, %ecx ; SSE2-SSSE3-NEXT: movl %eax, %ecx
; SSE2-SSSE3-NEXT: shrl $4, %ecx ; SSE2-SSSE3-NEXT: shrb $4, %cl
; SSE2-SSSE3-NEXT: movq %rcx, %xmm0 ; SSE2-SSSE3-NEXT: movzbl %cl, %ecx
; SSE2-SSSE3-NEXT: andl $15, %eax ; SSE2-SSSE3-NEXT: andb $15, %al
; SSE2-SSSE3-NEXT: movq %rax, %xmm1 ; SSE2-SSSE3-NEXT: movzbl %al, %eax
; SSE2-SSSE3-NEXT: punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0] ; SSE2-SSSE3-NEXT: movd %eax, %xmm0
; SSE2-SSSE3-NEXT: movdqa %xmm1, -{{[0-9]+}}(%rsp) ; SSE2-SSSE3-NEXT: pinsrw $4, %ecx, %xmm0
; SSE2-SSSE3-NEXT: movdqa %xmm0, -{{[0-9]+}}(%rsp)
; SSE2-SSSE3-NEXT: movb -{{[0-9]+}}(%rsp), %al ; SSE2-SSSE3-NEXT: movb -{{[0-9]+}}(%rsp), %al
; SSE2-SSSE3-NEXT: addb -{{[0-9]+}}(%rsp), %al ; SSE2-SSSE3-NEXT: addb -{{[0-9]+}}(%rsp), %al
; SSE2-SSSE3-NEXT: retq ; SSE2-SSSE3-NEXT: retq
; ;
; AVX12-LABEL: bitcast_v8i32_to_v2i4: ; AVX12-LABEL: bitcast_v8i32_to_v2i4:
; AVX12: # %bb.0: ; AVX12: # %bb.0:
; AVX12-NEXT: vmovmskps %ymm0, %ecx ; AVX12-NEXT: vmovmskps %ymm0, %eax
; AVX12-NEXT: movl %ecx, %eax ; AVX12-NEXT: movl %eax, %ecx
; AVX12-NEXT: shrl $4, %eax ; AVX12-NEXT: shrb $4, %cl
; AVX12-NEXT: andl $15, %ecx ; AVX12-NEXT: andb $15, %al
; AVX12-NEXT: addb %cl, %al ; AVX12-NEXT: addb %cl, %al
; AVX12-NEXT: # kill: def $al killed $al killed $eax ; AVX12-NEXT: # kill: def $al killed $al killed $eax
; AVX12-NEXT: vzeroupper ; AVX12-NEXT: vzeroupper
; AVX12-NEXT: retq ; AVX12-NEXT: retq
; ;
; AVX512-LABEL: bitcast_v8i32_to_v2i4: ; AVX512-LABEL: bitcast_v8i32_to_v2i4:
; AVX512: # %bb.0: ; AVX512: # %bb.0:
; AVX512-NEXT: vpxor %xmm1, %xmm1, %xmm1 ; AVX512-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512-NEXT: vpcmpgtd %ymm0, %ymm1, %k0 ; AVX512-NEXT: vpcmpgtd %ymm0, %ymm1, %k0
; AVX512-NEXT: kmovd %k0, %ecx ; AVX512-NEXT: kmovd %k0, %eax
; AVX512-NEXT: movzbl %cl, %eax ; AVX512-NEXT: movl %eax, %ecx
; AVX512-NEXT: shrl $4, %eax ; AVX512-NEXT: shrb $4, %cl
; AVX512-NEXT: andl $15, %ecx ; AVX512-NEXT: andb $15, %al
; AVX512-NEXT: addb %cl, %al ; AVX512-NEXT: addb %cl, %al
; AVX512-NEXT: # kill: def $al killed $al killed $eax ; AVX512-NEXT: # kill: def $al killed $al killed $eax
; AVX512-NEXT: vzeroupper ; AVX512-NEXT: vzeroupper
; AVX512-NEXT: retq ; AVX512-NEXT: retq
%1 = icmp slt <8 x i32> %a0, zeroinitializer %1 = icmp slt <8 x i32> %a0, zeroinitializer
%2 = bitcast <8 x i1> %1 to <2 x i4> %2 = bitcast <8 x i1> %1 to <2 x i4>
%3 = extractelement <2 x i4> %2, i32 0 %3 = extractelement <2 x i4> %2, i32 0
%4 = extractelement <2 x i4> %2, i32 1 %4 = extractelement <2 x i4> %2, i32 1
▲ Show 20 Lines • Show All 116 Lines • ▼ Show 20 Lines
define i4 @bitcast_v8i64_to_v2i4(<8 x i64> %a0) nounwind { define i4 @bitcast_v8i64_to_v2i4(<8 x i64> %a0) nounwind {
; SSE2-SSSE3-LABEL: bitcast_v8i64_to_v2i4: ; SSE2-SSSE3-LABEL: bitcast_v8i64_to_v2i4:
; SSE2-SSSE3: # %bb.0: ; SSE2-SSSE3: # %bb.0:
; SSE2-SSSE3-NEXT: packssdw %xmm3, %xmm2 ; SSE2-SSSE3-NEXT: packssdw %xmm3, %xmm2
; SSE2-SSSE3-NEXT: packssdw %xmm1, %xmm0 ; SSE2-SSSE3-NEXT: packssdw %xmm1, %xmm0
; SSE2-SSSE3-NEXT: packssdw %xmm2, %xmm0 ; SSE2-SSSE3-NEXT: packssdw %xmm2, %xmm0
; SSE2-SSSE3-NEXT: packsswb %xmm0, %xmm0 ; SSE2-SSSE3-NEXT: packsswb %xmm0, %xmm0
; SSE2-SSSE3-NEXT: pmovmskb %xmm0, %eax ; SSE2-SSSE3-NEXT: pmovmskb %xmm0, %eax
; SSE2-SSSE3-NEXT: movzbl %al, %ecx ; SSE2-SSSE3-NEXT: movl %eax, %ecx
; SSE2-SSSE3-NEXT: shrl $4, %ecx ; SSE2-SSSE3-NEXT: shrb $4, %cl
; SSE2-SSSE3-NEXT: movq %rcx, %xmm0 ; SSE2-SSSE3-NEXT: movzbl %cl, %ecx
; SSE2-SSSE3-NEXT: andl $15, %eax ; SSE2-SSSE3-NEXT: andb $15, %al
; SSE2-SSSE3-NEXT: movq %rax, %xmm1 ; SSE2-SSSE3-NEXT: movzbl %al, %eax
; SSE2-SSSE3-NEXT: punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0] ; SSE2-SSSE3-NEXT: movd %eax, %xmm0
; SSE2-SSSE3-NEXT: movdqa %xmm1, -{{[0-9]+}}(%rsp) ; SSE2-SSSE3-NEXT: pinsrw $4, %ecx, %xmm0
; SSE2-SSSE3-NEXT: movdqa %xmm0, -{{[0-9]+}}(%rsp)
; SSE2-SSSE3-NEXT: movb -{{[0-9]+}}(%rsp), %al ; SSE2-SSSE3-NEXT: movb -{{[0-9]+}}(%rsp), %al
; SSE2-SSSE3-NEXT: addb -{{[0-9]+}}(%rsp), %al ; SSE2-SSSE3-NEXT: addb -{{[0-9]+}}(%rsp), %al
; SSE2-SSSE3-NEXT: retq ; SSE2-SSSE3-NEXT: retq
; ;
; AVX1-LABEL: bitcast_v8i64_to_v2i4: ; AVX1-LABEL: bitcast_v8i64_to_v2i4:
; AVX1: # %bb.0: ; AVX1: # %bb.0:
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm2 ; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm2
; AVX1-NEXT: vpxor %xmm3, %xmm3, %xmm3 ; AVX1-NEXT: vpxor %xmm3, %xmm3, %xmm3
; AVX1-NEXT: vpcmpgtq %xmm2, %xmm3, %xmm2 ; AVX1-NEXT: vpcmpgtq %xmm2, %xmm3, %xmm2
; AVX1-NEXT: vpcmpgtq %xmm0, %xmm3, %xmm0 ; AVX1-NEXT: vpcmpgtq %xmm0, %xmm3, %xmm0
; AVX1-NEXT: vpackssdw %xmm2, %xmm0, %xmm0 ; AVX1-NEXT: vpackssdw %xmm2, %xmm0, %xmm0
; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2 ; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2
; AVX1-NEXT: vpackssdw %xmm2, %xmm1, %xmm1 ; AVX1-NEXT: vpackssdw %xmm2, %xmm1, %xmm1
; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 ; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
; AVX1-NEXT: vmovmskps %ymm0, %ecx ; AVX1-NEXT: vmovmskps %ymm0, %eax
; AVX1-NEXT: movl %ecx, %eax ; AVX1-NEXT: movl %eax, %ecx
; AVX1-NEXT: shrl $4, %eax ; AVX1-NEXT: shrb $4, %cl
; AVX1-NEXT: andl $15, %ecx ; AVX1-NEXT: andb $15, %al
; AVX1-NEXT: addb %cl, %al ; AVX1-NEXT: addb %cl, %al
; AVX1-NEXT: # kill: def $al killed $al killed $eax ; AVX1-NEXT: # kill: def $al killed $al killed $eax
; AVX1-NEXT: vzeroupper ; AVX1-NEXT: vzeroupper
; AVX1-NEXT: retq ; AVX1-NEXT: retq
; ;
; AVX2-LABEL: bitcast_v8i64_to_v2i4: ; AVX2-LABEL: bitcast_v8i64_to_v2i4:
; AVX2: # %bb.0: ; AVX2: # %bb.0:
; AVX2-NEXT: vpackssdw %ymm1, %ymm0, %ymm0 ; AVX2-NEXT: vpackssdw %ymm1, %ymm0, %ymm0
; AVX2-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3] ; AVX2-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
; AVX2-NEXT: vmovmskps %ymm0, %ecx ; AVX2-NEXT: vmovmskps %ymm0, %eax
; AVX2-NEXT: movl %ecx, %eax ; AVX2-NEXT: movl %eax, %ecx
; AVX2-NEXT: shrl $4, %eax ; AVX2-NEXT: shrb $4, %cl
; AVX2-NEXT: andl $15, %ecx ; AVX2-NEXT: andb $15, %al
; AVX2-NEXT: addb %cl, %al ; AVX2-NEXT: addb %cl, %al
; AVX2-NEXT: # kill: def $al killed $al killed $eax ; AVX2-NEXT: # kill: def $al killed $al killed $eax
; AVX2-NEXT: vzeroupper ; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq ; AVX2-NEXT: retq
; ;
; AVX512-LABEL: bitcast_v8i64_to_v2i4: ; AVX512-LABEL: bitcast_v8i64_to_v2i4:
; AVX512: # %bb.0: ; AVX512: # %bb.0:
; AVX512-NEXT: vpxor %xmm1, %xmm1, %xmm1 ; AVX512-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512-NEXT: vpcmpgtq %zmm0, %zmm1, %k0 ; AVX512-NEXT: vpcmpgtq %zmm0, %zmm1, %k0
; AVX512-NEXT: kmovd %k0, %ecx ; AVX512-NEXT: kmovd %k0, %eax
; AVX512-NEXT: movzbl %cl, %eax ; AVX512-NEXT: movl %eax, %ecx
; AVX512-NEXT: shrl $4, %eax ; AVX512-NEXT: shrb $4, %cl
; AVX512-NEXT: andl $15, %ecx ; AVX512-NEXT: andb $15, %al
; AVX512-NEXT: addb %cl, %al ; AVX512-NEXT: addb %cl, %al
; AVX512-NEXT: # kill: def $al killed $al killed $eax ; AVX512-NEXT: # kill: def $al killed $al killed $eax
; AVX512-NEXT: vzeroupper ; AVX512-NEXT: vzeroupper
; AVX512-NEXT: retq ; AVX512-NEXT: retq
%1 = icmp slt <8 x i64> %a0, zeroinitializer %1 = icmp slt <8 x i64> %a0, zeroinitializer
%2 = bitcast <8 x i1> %1 to <2 x i4> %2 = bitcast <8 x i1> %1 to <2 x i4>
%3 = extractelement <2 x i4> %2, i32 0 %3 = extractelement <2 x i4> %2, i32 0
%4 = extractelement <2 x i4> %2, i32 1 %4 = extractelement <2 x i4> %2, i32 1
▲ Show 20 Lines • Show All 277 Lines • Show Last 20 Lines