[X86][SSE] Add tests for min/max partial reduction

As mentioned on D65047 / rL366933 the plan is to enable partial reduction handling wherever possible.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@368015 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/test/CodeGen/X86/horizontal-reduce-smax.ll b/test/CodeGen/X86/horizontal-reduce-smax.ll
index c08f08f383bc12251dabe2c2c785986598b65234..32fee3a2e397a72e710c0393bc4c2a6adea8064d 100644 (file)
--- a/test/CodeGen/X86/horizontal-reduce-smax.ll
+++ b/test/CodeGen/X86/horizontal-reduce-smax.ll
@@ -1829,3 +1829,441 @@ define i8 @test_reduce_v64i8(<64 x i8> %a0) {
   %19 = extractelement <64 x i8> %18, i32 0
   ret i8 %19
 }
+
+;
+; Partial Vector Reductions
+;
+
+define i16 @test_reduce_v16i16_v8i16(<16 x i16> %a0) {
+; X86-SSE2-LABEL: test_reduce_v16i16_v8i16:
+; X86-SSE2:       ## %bb.0:
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X86-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
+; X86-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
+; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X86-SSE2-NEXT:    psrld $16, %xmm1
+; X86-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
+; X86-SSE2-NEXT:    movd %xmm1, %eax
+; X86-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X86-SSE2-NEXT:    retl
+;
+; X86-SSE42-LABEL: test_reduce_v16i16_v8i16:
+; X86-SSE42:       ## %bb.0:
+; X86-SSE42-NEXT:    pxor LCPI12_0, %xmm0
+; X86-SSE42-NEXT:    phminposuw %xmm0, %xmm0
+; X86-SSE42-NEXT:    movd %xmm0, %eax
+; X86-SSE42-NEXT:    xorl $32767, %eax ## imm = 0x7FFF
+; X86-SSE42-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X86-SSE42-NEXT:    retl
+;
+; X86-AVX-LABEL: test_reduce_v16i16_v8i16:
+; X86-AVX:       ## %bb.0:
+; X86-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X86-AVX-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; X86-AVX-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
+; X86-AVX-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vmovd %xmm0, %eax
+; X86-AVX-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X86-AVX-NEXT:    vzeroupper
+; X86-AVX-NEXT:    retl
+;
+; X64-SSE2-LABEL: test_reduce_v16i16_v8i16:
+; X64-SSE2:       ## %bb.0:
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X64-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
+; X64-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
+; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X64-SSE2-NEXT:    psrld $16, %xmm1
+; X64-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
+; X64-SSE2-NEXT:    movd %xmm1, %eax
+; X64-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X64-SSE2-NEXT:    retq
+;
+; X64-SSE42-LABEL: test_reduce_v16i16_v8i16:
+; X64-SSE42:       ## %bb.0:
+; X64-SSE42-NEXT:    pxor {{.*}}(%rip), %xmm0
+; X64-SSE42-NEXT:    phminposuw %xmm0, %xmm0
+; X64-SSE42-NEXT:    movd %xmm0, %eax
+; X64-SSE42-NEXT:    xorl $32767, %eax ## imm = 0x7FFF
+; X64-SSE42-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X64-SSE42-NEXT:    retq
+;
+; X64-AVX-LABEL: test_reduce_v16i16_v8i16:
+; X64-AVX:       ## %bb.0:
+; X64-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X64-AVX-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; X64-AVX-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
+; X64-AVX-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vmovd %xmm0, %eax
+; X64-AVX-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X64-AVX-NEXT:    vzeroupper
+; X64-AVX-NEXT:    retq
+  %1  = shufflevector <16 x i16> %a0, <16 x i16> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %2  = icmp sgt <16 x i16> %a0, %1
+  %3  = select <16 x i1> %2, <16 x i16> %a0, <16 x i16> %1
+  %4  = shufflevector <16 x i16> %3, <16 x i16> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %5  = icmp sgt <16 x i16> %3, %4
+  %6  = select <16 x i1> %5, <16 x i16> %3, <16 x i16> %4
+  %7  = shufflevector <16 x i16> %6, <16 x i16> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %8  = icmp sgt <16 x i16> %6, %7
+  %9  = select <16 x i1> %8, <16 x i16> %6, <16 x i16> %7
+  %10 = extractelement <16 x i16> %9, i32 0
+  ret i16 %10
+}
+
+define i16 @test_reduce_v32i16_v8i16(<32 x i16> %a0) {
+; X86-SSE2-LABEL: test_reduce_v32i16_v8i16:
+; X86-SSE2:       ## %bb.0:
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X86-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
+; X86-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
+; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X86-SSE2-NEXT:    psrld $16, %xmm1
+; X86-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
+; X86-SSE2-NEXT:    movd %xmm1, %eax
+; X86-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X86-SSE2-NEXT:    retl
+;
+; X86-SSE42-LABEL: test_reduce_v32i16_v8i16:
+; X86-SSE42:       ## %bb.0:
+; X86-SSE42-NEXT:    pxor LCPI13_0, %xmm0
+; X86-SSE42-NEXT:    phminposuw %xmm0, %xmm0
+; X86-SSE42-NEXT:    movd %xmm0, %eax
+; X86-SSE42-NEXT:    xorl $32767, %eax ## imm = 0x7FFF
+; X86-SSE42-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X86-SSE42-NEXT:    retl
+;
+; X86-AVX-LABEL: test_reduce_v32i16_v8i16:
+; X86-AVX:       ## %bb.0:
+; X86-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X86-AVX-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; X86-AVX-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
+; X86-AVX-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vmovd %xmm0, %eax
+; X86-AVX-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X86-AVX-NEXT:    vzeroupper
+; X86-AVX-NEXT:    retl
+;
+; X64-SSE2-LABEL: test_reduce_v32i16_v8i16:
+; X64-SSE2:       ## %bb.0:
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X64-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
+; X64-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
+; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X64-SSE2-NEXT:    psrld $16, %xmm1
+; X64-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
+; X64-SSE2-NEXT:    movd %xmm1, %eax
+; X64-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X64-SSE2-NEXT:    retq
+;
+; X64-SSE42-LABEL: test_reduce_v32i16_v8i16:
+; X64-SSE42:       ## %bb.0:
+; X64-SSE42-NEXT:    pxor {{.*}}(%rip), %xmm0
+; X64-SSE42-NEXT:    phminposuw %xmm0, %xmm0
+; X64-SSE42-NEXT:    movd %xmm0, %eax
+; X64-SSE42-NEXT:    xorl $32767, %eax ## imm = 0x7FFF
+; X64-SSE42-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X64-SSE42-NEXT:    retq
+;
+; X64-AVX-LABEL: test_reduce_v32i16_v8i16:
+; X64-AVX:       ## %bb.0:
+; X64-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X64-AVX-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; X64-AVX-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
+; X64-AVX-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vmovd %xmm0, %eax
+; X64-AVX-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X64-AVX-NEXT:    vzeroupper
+; X64-AVX-NEXT:    retq
+  %1  = shufflevector <32 x i16> %a0, <32 x i16> undef, <32 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %2  = icmp sgt <32 x i16> %a0, %1
+  %3  = select <32 x i1> %2, <32 x i16> %a0, <32 x i16> %1
+  %4  = shufflevector <32 x i16> %3, <32 x i16> undef, <32 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %5  = icmp sgt <32 x i16> %3, %4
+  %6  = select <32 x i1> %5, <32 x i16> %3, <32 x i16> %4
+  %7  = shufflevector <32 x i16> %6, <32 x i16> undef, <32 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %8  = icmp sgt <32 x i16> %6, %7
+  %9  = select <32 x i1> %8, <32 x i16> %6, <32 x i16> %7
+  %10 = extractelement <32 x i16> %9, i32 0
+  ret i16 %10
+}
+
+define i8 @test_reduce_v32i8_v16i8(<32 x i8> %a0) {
+; X86-SSE2-LABEL: test_reduce_v32i8_v16i8:
+; X86-SSE2:       ## %bb.0:
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X86-SSE2-NEXT:    movdqa %xmm0, %xmm2
+; X86-SSE2-NEXT:    pcmpgtb %xmm1, %xmm2
+; X86-SSE2-NEXT:    pand %xmm2, %xmm0
+; X86-SSE2-NEXT:    pandn %xmm1, %xmm2
+; X86-SSE2-NEXT:    por %xmm0, %xmm2
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[1,1,2,3]
+; X86-SSE2-NEXT:    movdqa %xmm2, %xmm1
+; X86-SSE2-NEXT:    pcmpgtb %xmm0, %xmm1
+; X86-SSE2-NEXT:    pand %xmm1, %xmm2
+; X86-SSE2-NEXT:    pandn %xmm0, %xmm1
+; X86-SSE2-NEXT:    por %xmm2, %xmm1
+; X86-SSE2-NEXT:    movdqa %xmm1, %xmm0
+; X86-SSE2-NEXT:    psrld $16, %xmm0
+; X86-SSE2-NEXT:    movdqa %xmm1, %xmm2
+; X86-SSE2-NEXT:    pcmpgtb %xmm0, %xmm2
+; X86-SSE2-NEXT:    pand %xmm2, %xmm1
+; X86-SSE2-NEXT:    pandn %xmm0, %xmm2
+; X86-SSE2-NEXT:    por %xmm1, %xmm2
+; X86-SSE2-NEXT:    movdqa %xmm2, %xmm0
+; X86-SSE2-NEXT:    psrlw $8, %xmm0
+; X86-SSE2-NEXT:    movdqa %xmm2, %xmm1
+; X86-SSE2-NEXT:    pcmpgtb %xmm0, %xmm1
+; X86-SSE2-NEXT:    pand %xmm1, %xmm2
+; X86-SSE2-NEXT:    pandn %xmm0, %xmm1
+; X86-SSE2-NEXT:    por %xmm2, %xmm1
+; X86-SSE2-NEXT:    movd %xmm1, %eax
+; X86-SSE2-NEXT:    ## kill: def $al killed $al killed $eax
+; X86-SSE2-NEXT:    retl
+;
+; X86-SSE42-LABEL: test_reduce_v32i8_v16i8:
+; X86-SSE42:       ## %bb.0:
+; X86-SSE42-NEXT:    pxor LCPI14_0, %xmm0
+; X86-SSE42-NEXT:    movdqa %xmm0, %xmm1
+; X86-SSE42-NEXT:    psrlw $8, %xmm1
+; X86-SSE42-NEXT:    pminub %xmm0, %xmm1
+; X86-SSE42-NEXT:    phminposuw %xmm1, %xmm0
+; X86-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X86-SSE42-NEXT:    xorb $127, %al
+; X86-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
+; X86-SSE42-NEXT:    retl
+;
+; X86-AVX-LABEL: test_reduce_v32i8_v16i8:
+; X86-AVX:       ## %bb.0:
+; X86-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X86-AVX-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; X86-AVX-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
+; X86-AVX-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X86-AVX-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX-NEXT:    ## kill: def $al killed $al killed $eax
+; X86-AVX-NEXT:    vzeroupper
+; X86-AVX-NEXT:    retl
+;
+; X64-SSE2-LABEL: test_reduce_v32i8_v16i8:
+; X64-SSE2:       ## %bb.0:
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X64-SSE2-NEXT:    movdqa %xmm0, %xmm2
+; X64-SSE2-NEXT:    pcmpgtb %xmm1, %xmm2
+; X64-SSE2-NEXT:    pand %xmm2, %xmm0
+; X64-SSE2-NEXT:    pandn %xmm1, %xmm2
+; X64-SSE2-NEXT:    por %xmm0, %xmm2
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[1,1,2,3]
+; X64-SSE2-NEXT:    movdqa %xmm2, %xmm1
+; X64-SSE2-NEXT:    pcmpgtb %xmm0, %xmm1
+; X64-SSE2-NEXT:    pand %xmm1, %xmm2
+; X64-SSE2-NEXT:    pandn %xmm0, %xmm1
+; X64-SSE2-NEXT:    por %xmm2, %xmm1
+; X64-SSE2-NEXT:    movdqa %xmm1, %xmm0
+; X64-SSE2-NEXT:    psrld $16, %xmm0
+; X64-SSE2-NEXT:    movdqa %xmm1, %xmm2
+; X64-SSE2-NEXT:    pcmpgtb %xmm0, %xmm2
+; X64-SSE2-NEXT:    pand %xmm2, %xmm1
+; X64-SSE2-NEXT:    pandn %xmm0, %xmm2
+; X64-SSE2-NEXT:    por %xmm1, %xmm2
+; X64-SSE2-NEXT:    movdqa %xmm2, %xmm0
+; X64-SSE2-NEXT:    psrlw $8, %xmm0
+; X64-SSE2-NEXT:    movdqa %xmm2, %xmm1
+; X64-SSE2-NEXT:    pcmpgtb %xmm0, %xmm1
+; X64-SSE2-NEXT:    pand %xmm1, %xmm2
+; X64-SSE2-NEXT:    pandn %xmm0, %xmm1
+; X64-SSE2-NEXT:    por %xmm2, %xmm1
+; X64-SSE2-NEXT:    movd %xmm1, %eax
+; X64-SSE2-NEXT:    ## kill: def $al killed $al killed $eax
+; X64-SSE2-NEXT:    retq
+;
+; X64-SSE42-LABEL: test_reduce_v32i8_v16i8:
+; X64-SSE42:       ## %bb.0:
+; X64-SSE42-NEXT:    pxor {{.*}}(%rip), %xmm0
+; X64-SSE42-NEXT:    movdqa %xmm0, %xmm1
+; X64-SSE42-NEXT:    psrlw $8, %xmm1
+; X64-SSE42-NEXT:    pminub %xmm0, %xmm1
+; X64-SSE42-NEXT:    phminposuw %xmm1, %xmm0
+; X64-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X64-SSE42-NEXT:    xorb $127, %al
+; X64-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
+; X64-SSE42-NEXT:    retq
+;
+; X64-AVX-LABEL: test_reduce_v32i8_v16i8:
+; X64-AVX:       ## %bb.0:
+; X64-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X64-AVX-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; X64-AVX-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
+; X64-AVX-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X64-AVX-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX-NEXT:    ## kill: def $al killed $al killed $eax
+; X64-AVX-NEXT:    vzeroupper
+; X64-AVX-NEXT:    retq
+  %1  = shufflevector <32 x i8> %a0, <32 x i8> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %2  = icmp sgt <32 x i8> %a0, %1
+  %3  = select <32 x i1> %2, <32 x i8> %a0, <32 x i8> %1
+  %4  = shufflevector <32 x i8> %3, <32 x i8> undef, <32 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %5  = icmp sgt <32 x i8> %3, %4
+  %6  = select <32 x i1> %5, <32 x i8> %3, <32 x i8> %4
+  %7  = shufflevector <32 x i8> %6, <32 x i8> undef, <32 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %8  = icmp sgt <32 x i8> %6, %7
+  %9  = select <32 x i1> %8, <32 x i8> %6, <32 x i8> %7
+  %10 = shufflevector <32 x i8> %9, <32 x i8> undef, <32 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %11 = icmp sgt <32 x i8> %9, %10
+  %12 = select <32 x i1> %11, <32 x i8> %9, <32 x i8> %10
+  %13 = extractelement <32 x i8> %12, i32 0
+  ret i8 %13
+}
+
+define i8 @test_reduce_v64i8_v16i8(<64 x i8> %a0) {
+; X86-SSE2-LABEL: test_reduce_v64i8_v16i8:
+; X86-SSE2:       ## %bb.0:
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X86-SSE2-NEXT:    movdqa %xmm0, %xmm2
+; X86-SSE2-NEXT:    pcmpgtb %xmm1, %xmm2
+; X86-SSE2-NEXT:    pand %xmm2, %xmm0
+; X86-SSE2-NEXT:    pandn %xmm1, %xmm2
+; X86-SSE2-NEXT:    por %xmm0, %xmm2
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[1,1,2,3]
+; X86-SSE2-NEXT:    movdqa %xmm2, %xmm1
+; X86-SSE2-NEXT:    pcmpgtb %xmm0, %xmm1
+; X86-SSE2-NEXT:    pand %xmm1, %xmm2
+; X86-SSE2-NEXT:    pandn %xmm0, %xmm1
+; X86-SSE2-NEXT:    por %xmm2, %xmm1
+; X86-SSE2-NEXT:    movdqa %xmm1, %xmm0
+; X86-SSE2-NEXT:    psrld $16, %xmm0
+; X86-SSE2-NEXT:    movdqa %xmm1, %xmm2
+; X86-SSE2-NEXT:    pcmpgtb %xmm0, %xmm2
+; X86-SSE2-NEXT:    pand %xmm2, %xmm1
+; X86-SSE2-NEXT:    pandn %xmm0, %xmm2
+; X86-SSE2-NEXT:    por %xmm1, %xmm2
+; X86-SSE2-NEXT:    movdqa %xmm2, %xmm0
+; X86-SSE2-NEXT:    psrlw $8, %xmm0
+; X86-SSE2-NEXT:    movdqa %xmm2, %xmm1
+; X86-SSE2-NEXT:    pcmpgtb %xmm0, %xmm1
+; X86-SSE2-NEXT:    pand %xmm1, %xmm2
+; X86-SSE2-NEXT:    pandn %xmm0, %xmm1
+; X86-SSE2-NEXT:    por %xmm2, %xmm1
+; X86-SSE2-NEXT:    movd %xmm1, %eax
+; X86-SSE2-NEXT:    ## kill: def $al killed $al killed $eax
+; X86-SSE2-NEXT:    retl
+;
+; X86-SSE42-LABEL: test_reduce_v64i8_v16i8:
+; X86-SSE42:       ## %bb.0:
+; X86-SSE42-NEXT:    pxor LCPI15_0, %xmm0
+; X86-SSE42-NEXT:    movdqa %xmm0, %xmm1
+; X86-SSE42-NEXT:    psrlw $8, %xmm1
+; X86-SSE42-NEXT:    pminub %xmm0, %xmm1
+; X86-SSE42-NEXT:    phminposuw %xmm1, %xmm0
+; X86-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X86-SSE42-NEXT:    xorb $127, %al
+; X86-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
+; X86-SSE42-NEXT:    retl
+;
+; X86-AVX-LABEL: test_reduce_v64i8_v16i8:
+; X86-AVX:       ## %bb.0:
+; X86-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X86-AVX-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; X86-AVX-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
+; X86-AVX-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X86-AVX-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX-NEXT:    ## kill: def $al killed $al killed $eax
+; X86-AVX-NEXT:    vzeroupper
+; X86-AVX-NEXT:    retl
+;
+; X64-SSE2-LABEL: test_reduce_v64i8_v16i8:
+; X64-SSE2:       ## %bb.0:
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X64-SSE2-NEXT:    movdqa %xmm0, %xmm2
+; X64-SSE2-NEXT:    pcmpgtb %xmm1, %xmm2
+; X64-SSE2-NEXT:    pand %xmm2, %xmm0
+; X64-SSE2-NEXT:    pandn %xmm1, %xmm2
+; X64-SSE2-NEXT:    por %xmm0, %xmm2
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[1,1,2,3]
+; X64-SSE2-NEXT:    movdqa %xmm2, %xmm1
+; X64-SSE2-NEXT:    pcmpgtb %xmm0, %xmm1
+; X64-SSE2-NEXT:    pand %xmm1, %xmm2
+; X64-SSE2-NEXT:    pandn %xmm0, %xmm1
+; X64-SSE2-NEXT:    por %xmm2, %xmm1
+; X64-SSE2-NEXT:    movdqa %xmm1, %xmm0
+; X64-SSE2-NEXT:    psrld $16, %xmm0
+; X64-SSE2-NEXT:    movdqa %xmm1, %xmm2
+; X64-SSE2-NEXT:    pcmpgtb %xmm0, %xmm2
+; X64-SSE2-NEXT:    pand %xmm2, %xmm1
+; X64-SSE2-NEXT:    pandn %xmm0, %xmm2
+; X64-SSE2-NEXT:    por %xmm1, %xmm2
+; X64-SSE2-NEXT:    movdqa %xmm2, %xmm0
+; X64-SSE2-NEXT:    psrlw $8, %xmm0
+; X64-SSE2-NEXT:    movdqa %xmm2, %xmm1
+; X64-SSE2-NEXT:    pcmpgtb %xmm0, %xmm1
+; X64-SSE2-NEXT:    pand %xmm1, %xmm2
+; X64-SSE2-NEXT:    pandn %xmm0, %xmm1
+; X64-SSE2-NEXT:    por %xmm2, %xmm1
+; X64-SSE2-NEXT:    movd %xmm1, %eax
+; X64-SSE2-NEXT:    ## kill: def $al killed $al killed $eax
+; X64-SSE2-NEXT:    retq
+;
+; X64-SSE42-LABEL: test_reduce_v64i8_v16i8:
+; X64-SSE42:       ## %bb.0:
+; X64-SSE42-NEXT:    pxor {{.*}}(%rip), %xmm0
+; X64-SSE42-NEXT:    movdqa %xmm0, %xmm1
+; X64-SSE42-NEXT:    psrlw $8, %xmm1
+; X64-SSE42-NEXT:    pminub %xmm0, %xmm1
+; X64-SSE42-NEXT:    phminposuw %xmm1, %xmm0
+; X64-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X64-SSE42-NEXT:    xorb $127, %al
+; X64-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
+; X64-SSE42-NEXT:    retq
+;
+; X64-AVX-LABEL: test_reduce_v64i8_v16i8:
+; X64-AVX:       ## %bb.0:
+; X64-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X64-AVX-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; X64-AVX-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
+; X64-AVX-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X64-AVX-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX-NEXT:    ## kill: def $al killed $al killed $eax
+; X64-AVX-NEXT:    vzeroupper
+; X64-AVX-NEXT:    retq
+  %1  = shufflevector <64 x i8> %a0, <64 x i8> undef, <64 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %2  = icmp sgt <64 x i8> %a0, %1
+  %3  = select <64 x i1> %2, <64 x i8> %a0, <64 x i8> %1
+  %4  = shufflevector <64 x i8> %3, <64 x i8> undef, <64 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %5  = icmp sgt <64 x i8> %3, %4
+  %6  = select <64 x i1> %5, <64 x i8> %3, <64 x i8> %4
+  %7  = shufflevector <64 x i8> %6, <64 x i8> undef, <64 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %8  = icmp sgt <64 x i8> %6, %7
+  %9  = select <64 x i1> %8, <64 x i8> %6, <64 x i8> %7
+  %10 = shufflevector <64 x i8> %9, <64 x i8> undef, <64 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %11 = icmp sgt <64 x i8> %9, %10
+  %12 = select <64 x i1> %11, <64 x i8> %9, <64 x i8> %10
+  %13 = extractelement <64 x i8> %12, i32 0
+  ret i8 %13
+}

diff --git a/test/CodeGen/X86/horizontal-reduce-smin.ll b/test/CodeGen/X86/horizontal-reduce-smin.ll
index c526cb8f99c25c9c571e3ac86bd3052c4b5591dc..a34432237826920c5896036f58301447b0d36f4b 100644 (file)
--- a/test/CodeGen/X86/horizontal-reduce-smin.ll
+++ b/test/CodeGen/X86/horizontal-reduce-smin.ll
@@ -1833,3 +1833,441 @@ define i8 @test_reduce_v64i8(<64 x i8> %a0) {
   %19 = extractelement <64 x i8> %18, i32 0
   ret i8 %19
 }
+
+;
+; Partial Vector Reductions
+;
+
+define i16 @test_reduce_v16i16_v8i16(<16 x i16> %a0) {
+; X86-SSE2-LABEL: test_reduce_v16i16_v8i16:
+; X86-SSE2:       ## %bb.0:
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X86-SSE2-NEXT:    pminsw %xmm0, %xmm1
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
+; X86-SSE2-NEXT:    pminsw %xmm1, %xmm0
+; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X86-SSE2-NEXT:    psrld $16, %xmm1
+; X86-SSE2-NEXT:    pminsw %xmm0, %xmm1
+; X86-SSE2-NEXT:    movd %xmm1, %eax
+; X86-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X86-SSE2-NEXT:    retl
+;
+; X86-SSE42-LABEL: test_reduce_v16i16_v8i16:
+; X86-SSE42:       ## %bb.0:
+; X86-SSE42-NEXT:    pxor LCPI12_0, %xmm0
+; X86-SSE42-NEXT:    phminposuw %xmm0, %xmm0
+; X86-SSE42-NEXT:    movd %xmm0, %eax
+; X86-SSE42-NEXT:    xorl $32768, %eax ## imm = 0x8000
+; X86-SSE42-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X86-SSE42-NEXT:    retl
+;
+; X86-AVX-LABEL: test_reduce_v16i16_v8i16:
+; X86-AVX:       ## %bb.0:
+; X86-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X86-AVX-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; X86-AVX-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
+; X86-AVX-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vmovd %xmm0, %eax
+; X86-AVX-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X86-AVX-NEXT:    vzeroupper
+; X86-AVX-NEXT:    retl
+;
+; X64-SSE2-LABEL: test_reduce_v16i16_v8i16:
+; X64-SSE2:       ## %bb.0:
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X64-SSE2-NEXT:    pminsw %xmm0, %xmm1
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
+; X64-SSE2-NEXT:    pminsw %xmm1, %xmm0
+; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X64-SSE2-NEXT:    psrld $16, %xmm1
+; X64-SSE2-NEXT:    pminsw %xmm0, %xmm1
+; X64-SSE2-NEXT:    movd %xmm1, %eax
+; X64-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X64-SSE2-NEXT:    retq
+;
+; X64-SSE42-LABEL: test_reduce_v16i16_v8i16:
+; X64-SSE42:       ## %bb.0:
+; X64-SSE42-NEXT:    pxor {{.*}}(%rip), %xmm0
+; X64-SSE42-NEXT:    phminposuw %xmm0, %xmm0
+; X64-SSE42-NEXT:    movd %xmm0, %eax
+; X64-SSE42-NEXT:    xorl $32768, %eax ## imm = 0x8000
+; X64-SSE42-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X64-SSE42-NEXT:    retq
+;
+; X64-AVX-LABEL: test_reduce_v16i16_v8i16:
+; X64-AVX:       ## %bb.0:
+; X64-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X64-AVX-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; X64-AVX-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
+; X64-AVX-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vmovd %xmm0, %eax
+; X64-AVX-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X64-AVX-NEXT:    vzeroupper
+; X64-AVX-NEXT:    retq
+  %1  = shufflevector <16 x i16> %a0, <16 x i16> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %2  = icmp slt <16 x i16> %a0, %1
+  %3  = select <16 x i1> %2, <16 x i16> %a0, <16 x i16> %1
+  %4  = shufflevector <16 x i16> %3, <16 x i16> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %5  = icmp slt <16 x i16> %3, %4
+  %6  = select <16 x i1> %5, <16 x i16> %3, <16 x i16> %4
+  %7  = shufflevector <16 x i16> %6, <16 x i16> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %8  = icmp slt <16 x i16> %6, %7
+  %9  = select <16 x i1> %8, <16 x i16> %6, <16 x i16> %7
+  %10 = extractelement <16 x i16> %9, i32 0
+  ret i16 %10
+}
+
+define i16 @test_reduce_v32i16_v8i16(<32 x i16> %a0) {
+; X86-SSE2-LABEL: test_reduce_v32i16_v8i16:
+; X86-SSE2:       ## %bb.0:
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X86-SSE2-NEXT:    pminsw %xmm0, %xmm1
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
+; X86-SSE2-NEXT:    pminsw %xmm1, %xmm0
+; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X86-SSE2-NEXT:    psrld $16, %xmm1
+; X86-SSE2-NEXT:    pminsw %xmm0, %xmm1
+; X86-SSE2-NEXT:    movd %xmm1, %eax
+; X86-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X86-SSE2-NEXT:    retl
+;
+; X86-SSE42-LABEL: test_reduce_v32i16_v8i16:
+; X86-SSE42:       ## %bb.0:
+; X86-SSE42-NEXT:    pxor LCPI13_0, %xmm0
+; X86-SSE42-NEXT:    phminposuw %xmm0, %xmm0
+; X86-SSE42-NEXT:    movd %xmm0, %eax
+; X86-SSE42-NEXT:    xorl $32768, %eax ## imm = 0x8000
+; X86-SSE42-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X86-SSE42-NEXT:    retl
+;
+; X86-AVX-LABEL: test_reduce_v32i16_v8i16:
+; X86-AVX:       ## %bb.0:
+; X86-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X86-AVX-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; X86-AVX-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
+; X86-AVX-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vmovd %xmm0, %eax
+; X86-AVX-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X86-AVX-NEXT:    vzeroupper
+; X86-AVX-NEXT:    retl
+;
+; X64-SSE2-LABEL: test_reduce_v32i16_v8i16:
+; X64-SSE2:       ## %bb.0:
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X64-SSE2-NEXT:    pminsw %xmm0, %xmm1
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
+; X64-SSE2-NEXT:    pminsw %xmm1, %xmm0
+; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X64-SSE2-NEXT:    psrld $16, %xmm1
+; X64-SSE2-NEXT:    pminsw %xmm0, %xmm1
+; X64-SSE2-NEXT:    movd %xmm1, %eax
+; X64-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X64-SSE2-NEXT:    retq
+;
+; X64-SSE42-LABEL: test_reduce_v32i16_v8i16:
+; X64-SSE42:       ## %bb.0:
+; X64-SSE42-NEXT:    pxor {{.*}}(%rip), %xmm0
+; X64-SSE42-NEXT:    phminposuw %xmm0, %xmm0
+; X64-SSE42-NEXT:    movd %xmm0, %eax
+; X64-SSE42-NEXT:    xorl $32768, %eax ## imm = 0x8000
+; X64-SSE42-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X64-SSE42-NEXT:    retq
+;
+; X64-AVX-LABEL: test_reduce_v32i16_v8i16:
+; X64-AVX:       ## %bb.0:
+; X64-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X64-AVX-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; X64-AVX-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
+; X64-AVX-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vmovd %xmm0, %eax
+; X64-AVX-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X64-AVX-NEXT:    vzeroupper
+; X64-AVX-NEXT:    retq
+  %1  = shufflevector <32 x i16> %a0, <32 x i16> undef, <32 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %2  = icmp slt <32 x i16> %a0, %1
+  %3  = select <32 x i1> %2, <32 x i16> %a0, <32 x i16> %1
+  %4  = shufflevector <32 x i16> %3, <32 x i16> undef, <32 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %5  = icmp slt <32 x i16> %3, %4
+  %6  = select <32 x i1> %5, <32 x i16> %3, <32 x i16> %4
+  %7  = shufflevector <32 x i16> %6, <32 x i16> undef, <32 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %8  = icmp slt <32 x i16> %6, %7
+  %9  = select <32 x i1> %8, <32 x i16> %6, <32 x i16> %7
+  %10 = extractelement <32 x i16> %9, i32 0
+  ret i16 %10
+}
+
+define i8 @test_reduce_v32i8_v16i8(<32 x i8> %a0) {
+; X86-SSE2-LABEL: test_reduce_v32i8_v16i8:
+; X86-SSE2:       ## %bb.0:
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X86-SSE2-NEXT:    movdqa %xmm1, %xmm2
+; X86-SSE2-NEXT:    pcmpgtb %xmm0, %xmm2
+; X86-SSE2-NEXT:    pand %xmm2, %xmm0
+; X86-SSE2-NEXT:    pandn %xmm1, %xmm2
+; X86-SSE2-NEXT:    por %xmm0, %xmm2
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[1,1,2,3]
+; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X86-SSE2-NEXT:    pcmpgtb %xmm2, %xmm1
+; X86-SSE2-NEXT:    pand %xmm1, %xmm2
+; X86-SSE2-NEXT:    pandn %xmm0, %xmm1
+; X86-SSE2-NEXT:    por %xmm2, %xmm1
+; X86-SSE2-NEXT:    movdqa %xmm1, %xmm0
+; X86-SSE2-NEXT:    psrld $16, %xmm0
+; X86-SSE2-NEXT:    movdqa %xmm0, %xmm2
+; X86-SSE2-NEXT:    pcmpgtb %xmm1, %xmm2
+; X86-SSE2-NEXT:    pand %xmm2, %xmm1
+; X86-SSE2-NEXT:    pandn %xmm0, %xmm2
+; X86-SSE2-NEXT:    por %xmm1, %xmm2
+; X86-SSE2-NEXT:    movdqa %xmm2, %xmm0
+; X86-SSE2-NEXT:    psrlw $8, %xmm0
+; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X86-SSE2-NEXT:    pcmpgtb %xmm2, %xmm1
+; X86-SSE2-NEXT:    pand %xmm1, %xmm2
+; X86-SSE2-NEXT:    pandn %xmm0, %xmm1
+; X86-SSE2-NEXT:    por %xmm2, %xmm1
+; X86-SSE2-NEXT:    movd %xmm1, %eax
+; X86-SSE2-NEXT:    ## kill: def $al killed $al killed $eax
+; X86-SSE2-NEXT:    retl
+;
+; X86-SSE42-LABEL: test_reduce_v32i8_v16i8:
+; X86-SSE42:       ## %bb.0:
+; X86-SSE42-NEXT:    pxor LCPI14_0, %xmm0
+; X86-SSE42-NEXT:    movdqa %xmm0, %xmm1
+; X86-SSE42-NEXT:    psrlw $8, %xmm1
+; X86-SSE42-NEXT:    pminub %xmm0, %xmm1
+; X86-SSE42-NEXT:    phminposuw %xmm1, %xmm0
+; X86-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X86-SSE42-NEXT:    xorb $-128, %al
+; X86-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
+; X86-SSE42-NEXT:    retl
+;
+; X86-AVX-LABEL: test_reduce_v32i8_v16i8:
+; X86-AVX:       ## %bb.0:
+; X86-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X86-AVX-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; X86-AVX-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
+; X86-AVX-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X86-AVX-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX-NEXT:    ## kill: def $al killed $al killed $eax
+; X86-AVX-NEXT:    vzeroupper
+; X86-AVX-NEXT:    retl
+;
+; X64-SSE2-LABEL: test_reduce_v32i8_v16i8:
+; X64-SSE2:       ## %bb.0:
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X64-SSE2-NEXT:    movdqa %xmm1, %xmm2
+; X64-SSE2-NEXT:    pcmpgtb %xmm0, %xmm2
+; X64-SSE2-NEXT:    pand %xmm2, %xmm0
+; X64-SSE2-NEXT:    pandn %xmm1, %xmm2
+; X64-SSE2-NEXT:    por %xmm0, %xmm2
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[1,1,2,3]
+; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X64-SSE2-NEXT:    pcmpgtb %xmm2, %xmm1
+; X64-SSE2-NEXT:    pand %xmm1, %xmm2
+; X64-SSE2-NEXT:    pandn %xmm0, %xmm1
+; X64-SSE2-NEXT:    por %xmm2, %xmm1
+; X64-SSE2-NEXT:    movdqa %xmm1, %xmm0
+; X64-SSE2-NEXT:    psrld $16, %xmm0
+; X64-SSE2-NEXT:    movdqa %xmm0, %xmm2
+; X64-SSE2-NEXT:    pcmpgtb %xmm1, %xmm2
+; X64-SSE2-NEXT:    pand %xmm2, %xmm1
+; X64-SSE2-NEXT:    pandn %xmm0, %xmm2
+; X64-SSE2-NEXT:    por %xmm1, %xmm2
+; X64-SSE2-NEXT:    movdqa %xmm2, %xmm0
+; X64-SSE2-NEXT:    psrlw $8, %xmm0
+; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X64-SSE2-NEXT:    pcmpgtb %xmm2, %xmm1
+; X64-SSE2-NEXT:    pand %xmm1, %xmm2
+; X64-SSE2-NEXT:    pandn %xmm0, %xmm1
+; X64-SSE2-NEXT:    por %xmm2, %xmm1
+; X64-SSE2-NEXT:    movd %xmm1, %eax
+; X64-SSE2-NEXT:    ## kill: def $al killed $al killed $eax
+; X64-SSE2-NEXT:    retq
+;
+; X64-SSE42-LABEL: test_reduce_v32i8_v16i8:
+; X64-SSE42:       ## %bb.0:
+; X64-SSE42-NEXT:    pxor {{.*}}(%rip), %xmm0
+; X64-SSE42-NEXT:    movdqa %xmm0, %xmm1
+; X64-SSE42-NEXT:    psrlw $8, %xmm1
+; X64-SSE42-NEXT:    pminub %xmm0, %xmm1
+; X64-SSE42-NEXT:    phminposuw %xmm1, %xmm0
+; X64-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X64-SSE42-NEXT:    xorb $-128, %al
+; X64-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
+; X64-SSE42-NEXT:    retq
+;
+; X64-AVX-LABEL: test_reduce_v32i8_v16i8:
+; X64-AVX:       ## %bb.0:
+; X64-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X64-AVX-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; X64-AVX-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
+; X64-AVX-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X64-AVX-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX-NEXT:    ## kill: def $al killed $al killed $eax
+; X64-AVX-NEXT:    vzeroupper
+; X64-AVX-NEXT:    retq
+  %1  = shufflevector <32 x i8> %a0, <32 x i8> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %2  = icmp slt <32 x i8> %a0, %1
+  %3  = select <32 x i1> %2, <32 x i8> %a0, <32 x i8> %1
+  %4  = shufflevector <32 x i8> %3, <32 x i8> undef, <32 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %5  = icmp slt <32 x i8> %3, %4
+  %6  = select <32 x i1> %5, <32 x i8> %3, <32 x i8> %4
+  %7  = shufflevector <32 x i8> %6, <32 x i8> undef, <32 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %8  = icmp slt <32 x i8> %6, %7
+  %9  = select <32 x i1> %8, <32 x i8> %6, <32 x i8> %7
+  %10 = shufflevector <32 x i8> %9, <32 x i8> undef, <32 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %11 = icmp slt <32 x i8> %9, %10
+  %12 = select <32 x i1> %11, <32 x i8> %9, <32 x i8> %10
+  %13 = extractelement <32 x i8> %12, i32 0
+  ret i8 %13
+}
+
+define i8 @test_reduce_v64i8_v16i8(<64 x i8> %a0) {
+; X86-SSE2-LABEL: test_reduce_v64i8_v16i8:
+; X86-SSE2:       ## %bb.0:
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X86-SSE2-NEXT:    movdqa %xmm1, %xmm2
+; X86-SSE2-NEXT:    pcmpgtb %xmm0, %xmm2
+; X86-SSE2-NEXT:    pand %xmm2, %xmm0
+; X86-SSE2-NEXT:    pandn %xmm1, %xmm2
+; X86-SSE2-NEXT:    por %xmm0, %xmm2
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[1,1,2,3]
+; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X86-SSE2-NEXT:    pcmpgtb %xmm2, %xmm1
+; X86-SSE2-NEXT:    pand %xmm1, %xmm2
+; X86-SSE2-NEXT:    pandn %xmm0, %xmm1
+; X86-SSE2-NEXT:    por %xmm2, %xmm1
+; X86-SSE2-NEXT:    movdqa %xmm1, %xmm0
+; X86-SSE2-NEXT:    psrld $16, %xmm0
+; X86-SSE2-NEXT:    movdqa %xmm0, %xmm2
+; X86-SSE2-NEXT:    pcmpgtb %xmm1, %xmm2
+; X86-SSE2-NEXT:    pand %xmm2, %xmm1
+; X86-SSE2-NEXT:    pandn %xmm0, %xmm2
+; X86-SSE2-NEXT:    por %xmm1, %xmm2
+; X86-SSE2-NEXT:    movdqa %xmm2, %xmm0
+; X86-SSE2-NEXT:    psrlw $8, %xmm0
+; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X86-SSE2-NEXT:    pcmpgtb %xmm2, %xmm1
+; X86-SSE2-NEXT:    pand %xmm1, %xmm2
+; X86-SSE2-NEXT:    pandn %xmm0, %xmm1
+; X86-SSE2-NEXT:    por %xmm2, %xmm1
+; X86-SSE2-NEXT:    movd %xmm1, %eax
+; X86-SSE2-NEXT:    ## kill: def $al killed $al killed $eax
+; X86-SSE2-NEXT:    retl
+;
+; X86-SSE42-LABEL: test_reduce_v64i8_v16i8:
+; X86-SSE42:       ## %bb.0:
+; X86-SSE42-NEXT:    pxor LCPI15_0, %xmm0
+; X86-SSE42-NEXT:    movdqa %xmm0, %xmm1
+; X86-SSE42-NEXT:    psrlw $8, %xmm1
+; X86-SSE42-NEXT:    pminub %xmm0, %xmm1
+; X86-SSE42-NEXT:    phminposuw %xmm1, %xmm0
+; X86-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X86-SSE42-NEXT:    xorb $-128, %al
+; X86-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
+; X86-SSE42-NEXT:    retl
+;
+; X86-AVX-LABEL: test_reduce_v64i8_v16i8:
+; X86-AVX:       ## %bb.0:
+; X86-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X86-AVX-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; X86-AVX-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
+; X86-AVX-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X86-AVX-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX-NEXT:    ## kill: def $al killed $al killed $eax
+; X86-AVX-NEXT:    vzeroupper
+; X86-AVX-NEXT:    retl
+;
+; X64-SSE2-LABEL: test_reduce_v64i8_v16i8:
+; X64-SSE2:       ## %bb.0:
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X64-SSE2-NEXT:    movdqa %xmm1, %xmm2
+; X64-SSE2-NEXT:    pcmpgtb %xmm0, %xmm2
+; X64-SSE2-NEXT:    pand %xmm2, %xmm0
+; X64-SSE2-NEXT:    pandn %xmm1, %xmm2
+; X64-SSE2-NEXT:    por %xmm0, %xmm2
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[1,1,2,3]
+; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X64-SSE2-NEXT:    pcmpgtb %xmm2, %xmm1
+; X64-SSE2-NEXT:    pand %xmm1, %xmm2
+; X64-SSE2-NEXT:    pandn %xmm0, %xmm1
+; X64-SSE2-NEXT:    por %xmm2, %xmm1
+; X64-SSE2-NEXT:    movdqa %xmm1, %xmm0
+; X64-SSE2-NEXT:    psrld $16, %xmm0
+; X64-SSE2-NEXT:    movdqa %xmm0, %xmm2
+; X64-SSE2-NEXT:    pcmpgtb %xmm1, %xmm2
+; X64-SSE2-NEXT:    pand %xmm2, %xmm1
+; X64-SSE2-NEXT:    pandn %xmm0, %xmm2
+; X64-SSE2-NEXT:    por %xmm1, %xmm2
+; X64-SSE2-NEXT:    movdqa %xmm2, %xmm0
+; X64-SSE2-NEXT:    psrlw $8, %xmm0
+; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X64-SSE2-NEXT:    pcmpgtb %xmm2, %xmm1
+; X64-SSE2-NEXT:    pand %xmm1, %xmm2
+; X64-SSE2-NEXT:    pandn %xmm0, %xmm1
+; X64-SSE2-NEXT:    por %xmm2, %xmm1
+; X64-SSE2-NEXT:    movd %xmm1, %eax
+; X64-SSE2-NEXT:    ## kill: def $al killed $al killed $eax
+; X64-SSE2-NEXT:    retq
+;
+; X64-SSE42-LABEL: test_reduce_v64i8_v16i8:
+; X64-SSE42:       ## %bb.0:
+; X64-SSE42-NEXT:    pxor {{.*}}(%rip), %xmm0
+; X64-SSE42-NEXT:    movdqa %xmm0, %xmm1
+; X64-SSE42-NEXT:    psrlw $8, %xmm1
+; X64-SSE42-NEXT:    pminub %xmm0, %xmm1
+; X64-SSE42-NEXT:    phminposuw %xmm1, %xmm0
+; X64-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X64-SSE42-NEXT:    xorb $-128, %al
+; X64-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
+; X64-SSE42-NEXT:    retq
+;
+; X64-AVX-LABEL: test_reduce_v64i8_v16i8:
+; X64-AVX:       ## %bb.0:
+; X64-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X64-AVX-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; X64-AVX-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
+; X64-AVX-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X64-AVX-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX-NEXT:    ## kill: def $al killed $al killed $eax
+; X64-AVX-NEXT:    vzeroupper
+; X64-AVX-NEXT:    retq
+  %1  = shufflevector <64 x i8> %a0, <64 x i8> undef, <64 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %2  = icmp slt <64 x i8> %a0, %1
+  %3  = select <64 x i1> %2, <64 x i8> %a0, <64 x i8> %1
+  %4  = shufflevector <64 x i8> %3, <64 x i8> undef, <64 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %5  = icmp slt <64 x i8> %3, %4
+  %6  = select <64 x i1> %5, <64 x i8> %3, <64 x i8> %4
+  %7  = shufflevector <64 x i8> %6, <64 x i8> undef, <64 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %8  = icmp slt <64 x i8> %6, %7
+  %9  = select <64 x i1> %8, <64 x i8> %6, <64 x i8> %7
+  %10 = shufflevector <64 x i8> %9, <64 x i8> undef, <64 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %11 = icmp slt <64 x i8> %9, %10
+  %12 = select <64 x i1> %11, <64 x i8> %9, <64 x i8> %10
+  %13 = extractelement <64 x i8> %12, i32 0
+  ret i8 %13
+}

diff --git a/test/CodeGen/X86/horizontal-reduce-umax.ll b/test/CodeGen/X86/horizontal-reduce-umax.ll
index 99038d7f3c1f22a13804b5fca4f2bff2e4ddf47a..05cd44d38857a67cce25e4a72237f7768dc91e2b 100644 (file)
--- a/test/CodeGen/X86/horizontal-reduce-umax.ll
+++ b/test/CodeGen/X86/horizontal-reduce-umax.ll
@@ -2021,3 +2021,421 @@ define i8 @test_reduce_v64i8(<64 x i8> %a0) {
   %19 = extractelement <64 x i8> %18, i32 0
   ret i8 %19
 }
+
+;
+; Partial Vector Reductions
+;
+
+define i16 @test_reduce_v16i16_v8i16(<16 x i16> %a0) {
+; X86-SSE2-LABEL: test_reduce_v16i16_v8i16:
+; X86-SSE2:       ## %bb.0:
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X86-SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
+; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
+; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
+; X86-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
+; X86-SSE2-NEXT:    movdqa %xmm1, %xmm0
+; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
+; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
+; X86-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
+; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
+; X86-SSE2-NEXT:    psrld $16, %xmm1
+; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
+; X86-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
+; X86-SSE2-NEXT:    movd %xmm1, %eax
+; X86-SSE2-NEXT:    xorl $32768, %eax ## imm = 0x8000
+; X86-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X86-SSE2-NEXT:    retl
+;
+; X86-SSE42-LABEL: test_reduce_v16i16_v8i16:
+; X86-SSE42:       ## %bb.0:
+; X86-SSE42-NEXT:    pcmpeqd %xmm1, %xmm1
+; X86-SSE42-NEXT:    pxor %xmm0, %xmm1
+; X86-SSE42-NEXT:    phminposuw %xmm1, %xmm0
+; X86-SSE42-NEXT:    movd %xmm0, %eax
+; X86-SSE42-NEXT:    notl %eax
+; X86-SSE42-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X86-SSE42-NEXT:    retl
+;
+; X86-AVX-LABEL: test_reduce_v16i16_v8i16:
+; X86-AVX:       ## %bb.0:
+; X86-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X86-AVX-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; X86-AVX-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
+; X86-AVX-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vmovd %xmm0, %eax
+; X86-AVX-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X86-AVX-NEXT:    vzeroupper
+; X86-AVX-NEXT:    retl
+;
+; X64-SSE2-LABEL: test_reduce_v16i16_v8i16:
+; X64-SSE2:       ## %bb.0:
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X64-SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
+; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
+; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
+; X64-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
+; X64-SSE2-NEXT:    movdqa %xmm1, %xmm0
+; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
+; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
+; X64-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
+; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
+; X64-SSE2-NEXT:    psrld $16, %xmm1
+; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
+; X64-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
+; X64-SSE2-NEXT:    movd %xmm1, %eax
+; X64-SSE2-NEXT:    xorl $32768, %eax ## imm = 0x8000
+; X64-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X64-SSE2-NEXT:    retq
+;
+; X64-SSE42-LABEL: test_reduce_v16i16_v8i16:
+; X64-SSE42:       ## %bb.0:
+; X64-SSE42-NEXT:    pcmpeqd %xmm1, %xmm1
+; X64-SSE42-NEXT:    pxor %xmm0, %xmm1
+; X64-SSE42-NEXT:    phminposuw %xmm1, %xmm0
+; X64-SSE42-NEXT:    movd %xmm0, %eax
+; X64-SSE42-NEXT:    notl %eax
+; X64-SSE42-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X64-SSE42-NEXT:    retq
+;
+; X64-AVX-LABEL: test_reduce_v16i16_v8i16:
+; X64-AVX:       ## %bb.0:
+; X64-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X64-AVX-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; X64-AVX-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
+; X64-AVX-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vmovd %xmm0, %eax
+; X64-AVX-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X64-AVX-NEXT:    vzeroupper
+; X64-AVX-NEXT:    retq
+  %1  = shufflevector <16 x i16> %a0, <16 x i16> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %2  = icmp ugt <16 x i16> %a0, %1
+  %3  = select <16 x i1> %2, <16 x i16> %a0, <16 x i16> %1
+  %4  = shufflevector <16 x i16> %3, <16 x i16> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %5  = icmp ugt <16 x i16> %3, %4
+  %6  = select <16 x i1> %5, <16 x i16> %3, <16 x i16> %4
+  %7  = shufflevector <16 x i16> %6, <16 x i16> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %8  = icmp ugt <16 x i16> %6, %7
+  %9  = select <16 x i1> %8, <16 x i16> %6, <16 x i16> %7
+  %10 = extractelement <16 x i16> %9, i32 0
+  ret i16 %10
+}
+
+define i16 @test_reduce_v32i16_v8i16(<32 x i16> %a0) {
+; X86-SSE2-LABEL: test_reduce_v32i16_v8i16:
+; X86-SSE2:       ## %bb.0:
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X86-SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
+; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
+; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
+; X86-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
+; X86-SSE2-NEXT:    movdqa %xmm1, %xmm0
+; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
+; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
+; X86-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
+; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
+; X86-SSE2-NEXT:    psrld $16, %xmm1
+; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
+; X86-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
+; X86-SSE2-NEXT:    movd %xmm1, %eax
+; X86-SSE2-NEXT:    xorl $32768, %eax ## imm = 0x8000
+; X86-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X86-SSE2-NEXT:    retl
+;
+; X86-SSE42-LABEL: test_reduce_v32i16_v8i16:
+; X86-SSE42:       ## %bb.0:
+; X86-SSE42-NEXT:    pcmpeqd %xmm1, %xmm1
+; X86-SSE42-NEXT:    pxor %xmm0, %xmm1
+; X86-SSE42-NEXT:    phminposuw %xmm1, %xmm0
+; X86-SSE42-NEXT:    movd %xmm0, %eax
+; X86-SSE42-NEXT:    notl %eax
+; X86-SSE42-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X86-SSE42-NEXT:    retl
+;
+; X86-AVX-LABEL: test_reduce_v32i16_v8i16:
+; X86-AVX:       ## %bb.0:
+; X86-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X86-AVX-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; X86-AVX-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
+; X86-AVX-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vmovd %xmm0, %eax
+; X86-AVX-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X86-AVX-NEXT:    vzeroupper
+; X86-AVX-NEXT:    retl
+;
+; X64-SSE2-LABEL: test_reduce_v32i16_v8i16:
+; X64-SSE2:       ## %bb.0:
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X64-SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
+; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
+; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
+; X64-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
+; X64-SSE2-NEXT:    movdqa %xmm1, %xmm0
+; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
+; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
+; X64-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
+; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
+; X64-SSE2-NEXT:    psrld $16, %xmm1
+; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
+; X64-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
+; X64-SSE2-NEXT:    movd %xmm1, %eax
+; X64-SSE2-NEXT:    xorl $32768, %eax ## imm = 0x8000
+; X64-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X64-SSE2-NEXT:    retq
+;
+; X64-SSE42-LABEL: test_reduce_v32i16_v8i16:
+; X64-SSE42:       ## %bb.0:
+; X64-SSE42-NEXT:    pcmpeqd %xmm1, %xmm1
+; X64-SSE42-NEXT:    pxor %xmm0, %xmm1
+; X64-SSE42-NEXT:    phminposuw %xmm1, %xmm0
+; X64-SSE42-NEXT:    movd %xmm0, %eax
+; X64-SSE42-NEXT:    notl %eax
+; X64-SSE42-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X64-SSE42-NEXT:    retq
+;
+; X64-AVX-LABEL: test_reduce_v32i16_v8i16:
+; X64-AVX:       ## %bb.0:
+; X64-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X64-AVX-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; X64-AVX-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
+; X64-AVX-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vmovd %xmm0, %eax
+; X64-AVX-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X64-AVX-NEXT:    vzeroupper
+; X64-AVX-NEXT:    retq
+  %1  = shufflevector <32 x i16> %a0, <32 x i16> undef, <32 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %2  = icmp ugt <32 x i16> %a0, %1
+  %3  = select <32 x i1> %2, <32 x i16> %a0, <32 x i16> %1
+  %4  = shufflevector <32 x i16> %3, <32 x i16> undef, <32 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %5  = icmp ugt <32 x i16> %3, %4
+  %6  = select <32 x i1> %5, <32 x i16> %3, <32 x i16> %4
+  %7  = shufflevector <32 x i16> %6, <32 x i16> undef, <32 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %8  = icmp ugt <32 x i16> %6, %7
+  %9  = select <32 x i1> %8, <32 x i16> %6, <32 x i16> %7
+  %10 = extractelement <32 x i16> %9, i32 0
+  ret i16 %10
+}
+
+define i8 @test_reduce_v32i8_v16i8(<32 x i8> %a0) {
+; X86-SSE2-LABEL: test_reduce_v32i8_v16i8:
+; X86-SSE2:       ## %bb.0:
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X86-SSE2-NEXT:    pmaxub %xmm0, %xmm1
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
+; X86-SSE2-NEXT:    pmaxub %xmm1, %xmm0
+; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X86-SSE2-NEXT:    psrld $16, %xmm1
+; X86-SSE2-NEXT:    pmaxub %xmm0, %xmm1
+; X86-SSE2-NEXT:    movdqa %xmm1, %xmm0
+; X86-SSE2-NEXT:    psrlw $8, %xmm0
+; X86-SSE2-NEXT:    pmaxub %xmm1, %xmm0
+; X86-SSE2-NEXT:    movd %xmm0, %eax
+; X86-SSE2-NEXT:    ## kill: def $al killed $al killed $eax
+; X86-SSE2-NEXT:    retl
+;
+; X86-SSE42-LABEL: test_reduce_v32i8_v16i8:
+; X86-SSE42:       ## %bb.0:
+; X86-SSE42-NEXT:    pcmpeqd %xmm1, %xmm1
+; X86-SSE42-NEXT:    pxor %xmm0, %xmm1
+; X86-SSE42-NEXT:    movdqa %xmm1, %xmm0
+; X86-SSE42-NEXT:    psrlw $8, %xmm0
+; X86-SSE42-NEXT:    pminub %xmm1, %xmm0
+; X86-SSE42-NEXT:    phminposuw %xmm0, %xmm0
+; X86-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X86-SSE42-NEXT:    notb %al
+; X86-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
+; X86-SSE42-NEXT:    retl
+;
+; X86-AVX-LABEL: test_reduce_v32i8_v16i8:
+; X86-AVX:       ## %bb.0:
+; X86-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X86-AVX-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; X86-AVX-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
+; X86-AVX-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X86-AVX-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX-NEXT:    ## kill: def $al killed $al killed $eax
+; X86-AVX-NEXT:    vzeroupper
+; X86-AVX-NEXT:    retl
+;
+; X64-SSE2-LABEL: test_reduce_v32i8_v16i8:
+; X64-SSE2:       ## %bb.0:
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X64-SSE2-NEXT:    pmaxub %xmm0, %xmm1
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
+; X64-SSE2-NEXT:    pmaxub %xmm1, %xmm0
+; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X64-SSE2-NEXT:    psrld $16, %xmm1
+; X64-SSE2-NEXT:    pmaxub %xmm0, %xmm1
+; X64-SSE2-NEXT:    movdqa %xmm1, %xmm0
+; X64-SSE2-NEXT:    psrlw $8, %xmm0
+; X64-SSE2-NEXT:    pmaxub %xmm1, %xmm0
+; X64-SSE2-NEXT:    movd %xmm0, %eax
+; X64-SSE2-NEXT:    ## kill: def $al killed $al killed $eax
+; X64-SSE2-NEXT:    retq
+;
+; X64-SSE42-LABEL: test_reduce_v32i8_v16i8:
+; X64-SSE42:       ## %bb.0:
+; X64-SSE42-NEXT:    pcmpeqd %xmm1, %xmm1
+; X64-SSE42-NEXT:    pxor %xmm0, %xmm1
+; X64-SSE42-NEXT:    movdqa %xmm1, %xmm0
+; X64-SSE42-NEXT:    psrlw $8, %xmm0
+; X64-SSE42-NEXT:    pminub %xmm1, %xmm0
+; X64-SSE42-NEXT:    phminposuw %xmm0, %xmm0
+; X64-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X64-SSE42-NEXT:    notb %al
+; X64-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
+; X64-SSE42-NEXT:    retq
+;
+; X64-AVX-LABEL: test_reduce_v32i8_v16i8:
+; X64-AVX:       ## %bb.0:
+; X64-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X64-AVX-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; X64-AVX-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
+; X64-AVX-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X64-AVX-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX-NEXT:    ## kill: def $al killed $al killed $eax
+; X64-AVX-NEXT:    vzeroupper
+; X64-AVX-NEXT:    retq
+  %1  = shufflevector <32 x i8> %a0, <32 x i8> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %2  = icmp ugt <32 x i8> %a0, %1
+  %3  = select <32 x i1> %2, <32 x i8> %a0, <32 x i8> %1
+  %4  = shufflevector <32 x i8> %3, <32 x i8> undef, <32 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %5  = icmp ugt <32 x i8> %3, %4
+  %6  = select <32 x i1> %5, <32 x i8> %3, <32 x i8> %4
+  %7  = shufflevector <32 x i8> %6, <32 x i8> undef, <32 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %8  = icmp ugt <32 x i8> %6, %7
+  %9  = select <32 x i1> %8, <32 x i8> %6, <32 x i8> %7
+  %10 = shufflevector <32 x i8> %9, <32 x i8> undef, <32 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %11 = icmp ugt <32 x i8> %9, %10
+  %12 = select <32 x i1> %11, <32 x i8> %9, <32 x i8> %10
+  %13 = extractelement <32 x i8> %12, i32 0
+  ret i8 %13
+}
+
+define i8 @test_reduce_v64i8_v16i8(<64 x i8> %a0) {
+; X86-SSE2-LABEL: test_reduce_v64i8_v16i8:
+; X86-SSE2:       ## %bb.0:
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X86-SSE2-NEXT:    pmaxub %xmm0, %xmm1
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
+; X86-SSE2-NEXT:    pmaxub %xmm1, %xmm0
+; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X86-SSE2-NEXT:    psrld $16, %xmm1
+; X86-SSE2-NEXT:    pmaxub %xmm0, %xmm1
+; X86-SSE2-NEXT:    movdqa %xmm1, %xmm0
+; X86-SSE2-NEXT:    psrlw $8, %xmm0
+; X86-SSE2-NEXT:    pmaxub %xmm1, %xmm0
+; X86-SSE2-NEXT:    movd %xmm0, %eax
+; X86-SSE2-NEXT:    ## kill: def $al killed $al killed $eax
+; X86-SSE2-NEXT:    retl
+;
+; X86-SSE42-LABEL: test_reduce_v64i8_v16i8:
+; X86-SSE42:       ## %bb.0:
+; X86-SSE42-NEXT:    pcmpeqd %xmm1, %xmm1
+; X86-SSE42-NEXT:    pxor %xmm0, %xmm1
+; X86-SSE42-NEXT:    movdqa %xmm1, %xmm0
+; X86-SSE42-NEXT:    psrlw $8, %xmm0
+; X86-SSE42-NEXT:    pminub %xmm1, %xmm0
+; X86-SSE42-NEXT:    phminposuw %xmm0, %xmm0
+; X86-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X86-SSE42-NEXT:    notb %al
+; X86-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
+; X86-SSE42-NEXT:    retl
+;
+; X86-AVX-LABEL: test_reduce_v64i8_v16i8:
+; X86-AVX:       ## %bb.0:
+; X86-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X86-AVX-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; X86-AVX-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
+; X86-AVX-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X86-AVX-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX-NEXT:    ## kill: def $al killed $al killed $eax
+; X86-AVX-NEXT:    vzeroupper
+; X86-AVX-NEXT:    retl
+;
+; X64-SSE2-LABEL: test_reduce_v64i8_v16i8:
+; X64-SSE2:       ## %bb.0:
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X64-SSE2-NEXT:    pmaxub %xmm0, %xmm1
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
+; X64-SSE2-NEXT:    pmaxub %xmm1, %xmm0
+; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X64-SSE2-NEXT:    psrld $16, %xmm1
+; X64-SSE2-NEXT:    pmaxub %xmm0, %xmm1
+; X64-SSE2-NEXT:    movdqa %xmm1, %xmm0
+; X64-SSE2-NEXT:    psrlw $8, %xmm0
+; X64-SSE2-NEXT:    pmaxub %xmm1, %xmm0
+; X64-SSE2-NEXT:    movd %xmm0, %eax
+; X64-SSE2-NEXT:    ## kill: def $al killed $al killed $eax
+; X64-SSE2-NEXT:    retq
+;
+; X64-SSE42-LABEL: test_reduce_v64i8_v16i8:
+; X64-SSE42:       ## %bb.0:
+; X64-SSE42-NEXT:    pcmpeqd %xmm1, %xmm1
+; X64-SSE42-NEXT:    pxor %xmm0, %xmm1
+; X64-SSE42-NEXT:    movdqa %xmm1, %xmm0
+; X64-SSE42-NEXT:    psrlw $8, %xmm0
+; X64-SSE42-NEXT:    pminub %xmm1, %xmm0
+; X64-SSE42-NEXT:    phminposuw %xmm0, %xmm0
+; X64-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X64-SSE42-NEXT:    notb %al
+; X64-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
+; X64-SSE42-NEXT:    retq
+;
+; X64-AVX-LABEL: test_reduce_v64i8_v16i8:
+; X64-AVX:       ## %bb.0:
+; X64-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X64-AVX-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; X64-AVX-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
+; X64-AVX-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X64-AVX-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX-NEXT:    ## kill: def $al killed $al killed $eax
+; X64-AVX-NEXT:    vzeroupper
+; X64-AVX-NEXT:    retq
+  %1  = shufflevector <64 x i8> %a0, <64 x i8> undef, <64 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %2  = icmp ugt <64 x i8> %a0, %1
+  %3  = select <64 x i1> %2, <64 x i8> %a0, <64 x i8> %1
+  %4  = shufflevector <64 x i8> %3, <64 x i8> undef, <64 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %5  = icmp ugt <64 x i8> %3, %4
+  %6  = select <64 x i1> %5, <64 x i8> %3, <64 x i8> %4
+  %7  = shufflevector <64 x i8> %6, <64 x i8> undef, <64 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %8  = icmp ugt <64 x i8> %6, %7
+  %9  = select <64 x i1> %8, <64 x i8> %6, <64 x i8> %7
+  %10 = shufflevector <64 x i8> %9, <64 x i8> undef, <64 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %11 = icmp ugt <64 x i8> %9, %10
+  %12 = select <64 x i1> %11, <64 x i8> %9, <64 x i8> %10
+  %13 = extractelement <64 x i8> %12, i32 0
+  ret i8 %13
+}

diff --git a/test/CodeGen/X86/horizontal-reduce-umin.ll b/test/CodeGen/X86/horizontal-reduce-umin.ll
index 5e4d83046d71f18a4b9c8ea76aa518d55bea58f9..ed64ec2eb27747a452c37dccc0b12a2aa3182f20 100644 (file)
--- a/test/CodeGen/X86/horizontal-reduce-umin.ll
+++ b/test/CodeGen/X86/horizontal-reduce-umin.ll
@@ -1885,3 +1885,397 @@ define i8 @test_reduce_v64i8(<64 x i8> %a0) {
   %19 = extractelement <64 x i8> %18, i32 0
   ret i8 %19
 }
+
+;
+; Partial Vector Reductions
+;
+
+define i16 @test_reduce_v16i16_v8i16(<16 x i16> %a0) {
+; X86-SSE2-LABEL: test_reduce_v16i16_v8i16:
+; X86-SSE2:       ## %bb.0:
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X86-SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
+; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
+; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
+; X86-SSE2-NEXT:    pminsw %xmm0, %xmm1
+; X86-SSE2-NEXT:    movdqa %xmm1, %xmm0
+; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
+; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
+; X86-SSE2-NEXT:    pminsw %xmm1, %xmm0
+; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
+; X86-SSE2-NEXT:    psrld $16, %xmm1
+; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
+; X86-SSE2-NEXT:    pminsw %xmm0, %xmm1
+; X86-SSE2-NEXT:    movd %xmm1, %eax
+; X86-SSE2-NEXT:    xorl $32768, %eax ## imm = 0x8000
+; X86-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X86-SSE2-NEXT:    retl
+;
+; X86-SSE42-LABEL: test_reduce_v16i16_v8i16:
+; X86-SSE42:       ## %bb.0:
+; X86-SSE42-NEXT:    phminposuw %xmm0, %xmm0
+; X86-SSE42-NEXT:    movd %xmm0, %eax
+; X86-SSE42-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X86-SSE42-NEXT:    retl
+;
+; X86-AVX-LABEL: test_reduce_v16i16_v8i16:
+; X86-AVX:       ## %bb.0:
+; X86-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X86-AVX-NEXT:    vpminuw %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; X86-AVX-NEXT:    vpminuw %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
+; X86-AVX-NEXT:    vpminuw %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vmovd %xmm0, %eax
+; X86-AVX-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X86-AVX-NEXT:    vzeroupper
+; X86-AVX-NEXT:    retl
+;
+; X64-SSE2-LABEL: test_reduce_v16i16_v8i16:
+; X64-SSE2:       ## %bb.0:
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X64-SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
+; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
+; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
+; X64-SSE2-NEXT:    pminsw %xmm0, %xmm1
+; X64-SSE2-NEXT:    movdqa %xmm1, %xmm0
+; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
+; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
+; X64-SSE2-NEXT:    pminsw %xmm1, %xmm0
+; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
+; X64-SSE2-NEXT:    psrld $16, %xmm1
+; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
+; X64-SSE2-NEXT:    pminsw %xmm0, %xmm1
+; X64-SSE2-NEXT:    movd %xmm1, %eax
+; X64-SSE2-NEXT:    xorl $32768, %eax ## imm = 0x8000
+; X64-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X64-SSE2-NEXT:    retq
+;
+; X64-SSE42-LABEL: test_reduce_v16i16_v8i16:
+; X64-SSE42:       ## %bb.0:
+; X64-SSE42-NEXT:    phminposuw %xmm0, %xmm0
+; X64-SSE42-NEXT:    movd %xmm0, %eax
+; X64-SSE42-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X64-SSE42-NEXT:    retq
+;
+; X64-AVX-LABEL: test_reduce_v16i16_v8i16:
+; X64-AVX:       ## %bb.0:
+; X64-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X64-AVX-NEXT:    vpminuw %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; X64-AVX-NEXT:    vpminuw %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
+; X64-AVX-NEXT:    vpminuw %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vmovd %xmm0, %eax
+; X64-AVX-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X64-AVX-NEXT:    vzeroupper
+; X64-AVX-NEXT:    retq
+  %1  = shufflevector <16 x i16> %a0, <16 x i16> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %2  = icmp ult <16 x i16> %a0, %1
+  %3  = select <16 x i1> %2, <16 x i16> %a0, <16 x i16> %1
+  %4  = shufflevector <16 x i16> %3, <16 x i16> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %5  = icmp ult <16 x i16> %3, %4
+  %6  = select <16 x i1> %5, <16 x i16> %3, <16 x i16> %4
+  %7  = shufflevector <16 x i16> %6, <16 x i16> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %8  = icmp ult <16 x i16> %6, %7
+  %9  = select <16 x i1> %8, <16 x i16> %6, <16 x i16> %7
+  %10 = extractelement <16 x i16> %9, i32 0
+  ret i16 %10
+}
+
+define i16 @test_reduce_v32i16_v8i16(<32 x i16> %a0) {
+; X86-SSE2-LABEL: test_reduce_v32i16_v8i16:
+; X86-SSE2:       ## %bb.0:
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X86-SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
+; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
+; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
+; X86-SSE2-NEXT:    pminsw %xmm0, %xmm1
+; X86-SSE2-NEXT:    movdqa %xmm1, %xmm0
+; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
+; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
+; X86-SSE2-NEXT:    pminsw %xmm1, %xmm0
+; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
+; X86-SSE2-NEXT:    psrld $16, %xmm1
+; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
+; X86-SSE2-NEXT:    pminsw %xmm0, %xmm1
+; X86-SSE2-NEXT:    movd %xmm1, %eax
+; X86-SSE2-NEXT:    xorl $32768, %eax ## imm = 0x8000
+; X86-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X86-SSE2-NEXT:    retl
+;
+; X86-SSE42-LABEL: test_reduce_v32i16_v8i16:
+; X86-SSE42:       ## %bb.0:
+; X86-SSE42-NEXT:    phminposuw %xmm0, %xmm0
+; X86-SSE42-NEXT:    movd %xmm0, %eax
+; X86-SSE42-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X86-SSE42-NEXT:    retl
+;
+; X86-AVX-LABEL: test_reduce_v32i16_v8i16:
+; X86-AVX:       ## %bb.0:
+; X86-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X86-AVX-NEXT:    vpminuw %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; X86-AVX-NEXT:    vpminuw %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
+; X86-AVX-NEXT:    vpminuw %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vmovd %xmm0, %eax
+; X86-AVX-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X86-AVX-NEXT:    vzeroupper
+; X86-AVX-NEXT:    retl
+;
+; X64-SSE2-LABEL: test_reduce_v32i16_v8i16:
+; X64-SSE2:       ## %bb.0:
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X64-SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
+; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
+; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
+; X64-SSE2-NEXT:    pminsw %xmm0, %xmm1
+; X64-SSE2-NEXT:    movdqa %xmm1, %xmm0
+; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
+; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
+; X64-SSE2-NEXT:    pminsw %xmm1, %xmm0
+; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
+; X64-SSE2-NEXT:    psrld $16, %xmm1
+; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
+; X64-SSE2-NEXT:    pminsw %xmm0, %xmm1
+; X64-SSE2-NEXT:    movd %xmm1, %eax
+; X64-SSE2-NEXT:    xorl $32768, %eax ## imm = 0x8000
+; X64-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X64-SSE2-NEXT:    retq
+;
+; X64-SSE42-LABEL: test_reduce_v32i16_v8i16:
+; X64-SSE42:       ## %bb.0:
+; X64-SSE42-NEXT:    phminposuw %xmm0, %xmm0
+; X64-SSE42-NEXT:    movd %xmm0, %eax
+; X64-SSE42-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X64-SSE42-NEXT:    retq
+;
+; X64-AVX-LABEL: test_reduce_v32i16_v8i16:
+; X64-AVX:       ## %bb.0:
+; X64-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X64-AVX-NEXT:    vpminuw %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; X64-AVX-NEXT:    vpminuw %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
+; X64-AVX-NEXT:    vpminuw %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vmovd %xmm0, %eax
+; X64-AVX-NEXT:    ## kill: def $ax killed $ax killed $eax
+; X64-AVX-NEXT:    vzeroupper
+; X64-AVX-NEXT:    retq
+  %1  = shufflevector <32 x i16> %a0, <32 x i16> undef, <32 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %2  = icmp ult <32 x i16> %a0, %1
+  %3  = select <32 x i1> %2, <32 x i16> %a0, <32 x i16> %1
+  %4  = shufflevector <32 x i16> %3, <32 x i16> undef, <32 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %5  = icmp ult <32 x i16> %3, %4
+  %6  = select <32 x i1> %5, <32 x i16> %3, <32 x i16> %4
+  %7  = shufflevector <32 x i16> %6, <32 x i16> undef, <32 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %8  = icmp ult <32 x i16> %6, %7
+  %9  = select <32 x i1> %8, <32 x i16> %6, <32 x i16> %7
+  %10 = extractelement <32 x i16> %9, i32 0
+  ret i16 %10
+}
+
+define i8 @test_reduce_v32i8_v16i8(<32 x i8> %a0) {
+; X86-SSE2-LABEL: test_reduce_v32i8_v16i8:
+; X86-SSE2:       ## %bb.0:
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X86-SSE2-NEXT:    pminub %xmm0, %xmm1
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
+; X86-SSE2-NEXT:    pminub %xmm1, %xmm0
+; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X86-SSE2-NEXT:    psrld $16, %xmm1
+; X86-SSE2-NEXT:    pminub %xmm0, %xmm1
+; X86-SSE2-NEXT:    movdqa %xmm1, %xmm0
+; X86-SSE2-NEXT:    psrlw $8, %xmm0
+; X86-SSE2-NEXT:    pminub %xmm1, %xmm0
+; X86-SSE2-NEXT:    movd %xmm0, %eax
+; X86-SSE2-NEXT:    ## kill: def $al killed $al killed $eax
+; X86-SSE2-NEXT:    retl
+;
+; X86-SSE42-LABEL: test_reduce_v32i8_v16i8:
+; X86-SSE42:       ## %bb.0:
+; X86-SSE42-NEXT:    movdqa %xmm0, %xmm1
+; X86-SSE42-NEXT:    psrlw $8, %xmm1
+; X86-SSE42-NEXT:    pminub %xmm0, %xmm1
+; X86-SSE42-NEXT:    phminposuw %xmm1, %xmm0
+; X86-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X86-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
+; X86-SSE42-NEXT:    retl
+;
+; X86-AVX-LABEL: test_reduce_v32i8_v16i8:
+; X86-AVX:       ## %bb.0:
+; X86-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X86-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; X86-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
+; X86-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X86-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX-NEXT:    ## kill: def $al killed $al killed $eax
+; X86-AVX-NEXT:    vzeroupper
+; X86-AVX-NEXT:    retl
+;
+; X64-SSE2-LABEL: test_reduce_v32i8_v16i8:
+; X64-SSE2:       ## %bb.0:
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X64-SSE2-NEXT:    pminub %xmm0, %xmm1
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
+; X64-SSE2-NEXT:    pminub %xmm1, %xmm0
+; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X64-SSE2-NEXT:    psrld $16, %xmm1
+; X64-SSE2-NEXT:    pminub %xmm0, %xmm1
+; X64-SSE2-NEXT:    movdqa %xmm1, %xmm0
+; X64-SSE2-NEXT:    psrlw $8, %xmm0
+; X64-SSE2-NEXT:    pminub %xmm1, %xmm0
+; X64-SSE2-NEXT:    movd %xmm0, %eax
+; X64-SSE2-NEXT:    ## kill: def $al killed $al killed $eax
+; X64-SSE2-NEXT:    retq
+;
+; X64-SSE42-LABEL: test_reduce_v32i8_v16i8:
+; X64-SSE42:       ## %bb.0:
+; X64-SSE42-NEXT:    movdqa %xmm0, %xmm1
+; X64-SSE42-NEXT:    psrlw $8, %xmm1
+; X64-SSE42-NEXT:    pminub %xmm0, %xmm1
+; X64-SSE42-NEXT:    phminposuw %xmm1, %xmm0
+; X64-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X64-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
+; X64-SSE42-NEXT:    retq
+;
+; X64-AVX-LABEL: test_reduce_v32i8_v16i8:
+; X64-AVX:       ## %bb.0:
+; X64-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X64-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; X64-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
+; X64-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X64-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX-NEXT:    ## kill: def $al killed $al killed $eax
+; X64-AVX-NEXT:    vzeroupper
+; X64-AVX-NEXT:    retq
+  %1  = shufflevector <32 x i8> %a0, <32 x i8> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %2  = icmp ult <32 x i8> %a0, %1
+  %3  = select <32 x i1> %2, <32 x i8> %a0, <32 x i8> %1
+  %4  = shufflevector <32 x i8> %3, <32 x i8> undef, <32 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %5  = icmp ult <32 x i8> %3, %4
+  %6  = select <32 x i1> %5, <32 x i8> %3, <32 x i8> %4
+  %7  = shufflevector <32 x i8> %6, <32 x i8> undef, <32 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %8  = icmp ult <32 x i8> %6, %7
+  %9  = select <32 x i1> %8, <32 x i8> %6, <32 x i8> %7
+  %10 = shufflevector <32 x i8> %9, <32 x i8> undef, <32 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %11 = icmp ult <32 x i8> %9, %10
+  %12 = select <32 x i1> %11, <32 x i8> %9, <32 x i8> %10
+  %13 = extractelement <32 x i8> %12, i32 0
+  ret i8 %13
+}
+
+define i8 @test_reduce_v64i8_v16i8(<64 x i8> %a0) {
+; X86-SSE2-LABEL: test_reduce_v64i8_v16i8:
+; X86-SSE2:       ## %bb.0:
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X86-SSE2-NEXT:    pminub %xmm0, %xmm1
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
+; X86-SSE2-NEXT:    pminub %xmm1, %xmm0
+; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X86-SSE2-NEXT:    psrld $16, %xmm1
+; X86-SSE2-NEXT:    pminub %xmm0, %xmm1
+; X86-SSE2-NEXT:    movdqa %xmm1, %xmm0
+; X86-SSE2-NEXT:    psrlw $8, %xmm0
+; X86-SSE2-NEXT:    pminub %xmm1, %xmm0
+; X86-SSE2-NEXT:    movd %xmm0, %eax
+; X86-SSE2-NEXT:    ## kill: def $al killed $al killed $eax
+; X86-SSE2-NEXT:    retl
+;
+; X86-SSE42-LABEL: test_reduce_v64i8_v16i8:
+; X86-SSE42:       ## %bb.0:
+; X86-SSE42-NEXT:    movdqa %xmm0, %xmm1
+; X86-SSE42-NEXT:    psrlw $8, %xmm1
+; X86-SSE42-NEXT:    pminub %xmm0, %xmm1
+; X86-SSE42-NEXT:    phminposuw %xmm1, %xmm0
+; X86-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X86-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
+; X86-SSE42-NEXT:    retl
+;
+; X86-AVX-LABEL: test_reduce_v64i8_v16i8:
+; X86-AVX:       ## %bb.0:
+; X86-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X86-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; X86-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
+; X86-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X86-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX-NEXT:    ## kill: def $al killed $al killed $eax
+; X86-AVX-NEXT:    vzeroupper
+; X86-AVX-NEXT:    retl
+;
+; X64-SSE2-LABEL: test_reduce_v64i8_v16i8:
+; X64-SSE2:       ## %bb.0:
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X64-SSE2-NEXT:    pminub %xmm0, %xmm1
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
+; X64-SSE2-NEXT:    pminub %xmm1, %xmm0
+; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X64-SSE2-NEXT:    psrld $16, %xmm1
+; X64-SSE2-NEXT:    pminub %xmm0, %xmm1
+; X64-SSE2-NEXT:    movdqa %xmm1, %xmm0
+; X64-SSE2-NEXT:    psrlw $8, %xmm0
+; X64-SSE2-NEXT:    pminub %xmm1, %xmm0
+; X64-SSE2-NEXT:    movd %xmm0, %eax
+; X64-SSE2-NEXT:    ## kill: def $al killed $al killed $eax
+; X64-SSE2-NEXT:    retq
+;
+; X64-SSE42-LABEL: test_reduce_v64i8_v16i8:
+; X64-SSE42:       ## %bb.0:
+; X64-SSE42-NEXT:    movdqa %xmm0, %xmm1
+; X64-SSE42-NEXT:    psrlw $8, %xmm1
+; X64-SSE42-NEXT:    pminub %xmm0, %xmm1
+; X64-SSE42-NEXT:    phminposuw %xmm1, %xmm0
+; X64-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X64-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
+; X64-SSE42-NEXT:    retq
+;
+; X64-AVX-LABEL: test_reduce_v64i8_v16i8:
+; X64-AVX:       ## %bb.0:
+; X64-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; X64-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; X64-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
+; X64-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X64-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX-NEXT:    ## kill: def $al killed $al killed $eax
+; X64-AVX-NEXT:    vzeroupper
+; X64-AVX-NEXT:    retq
+  %1  = shufflevector <64 x i8> %a0, <64 x i8> undef, <64 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %2  = icmp ult <64 x i8> %a0, %1
+  %3  = select <64 x i1> %2, <64 x i8> %a0, <64 x i8> %1
+  %4  = shufflevector <64 x i8> %3, <64 x i8> undef, <64 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %5  = icmp ult <64 x i8> %3, %4
+  %6  = select <64 x i1> %5, <64 x i8> %3, <64 x i8> %4
+  %7  = shufflevector <64 x i8> %6, <64 x i8> undef, <64 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %8  = icmp ult <64 x i8> %6, %7
+  %9  = select <64 x i1> %8, <64 x i8> %6, <64 x i8> %7
+  %10 = shufflevector <64 x i8> %9, <64 x i8> undef, <64 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %11 = icmp ult <64 x i8> %9, %10
+  %12 = select <64 x i1> %11, <64 x i8> %9, <64 x i8> %10
+  %13 = extractelement <64 x i8> %12, i32 0
+  ret i8 %13
+}