[x86] add tests for unpack shuffle lowering; NFC

https://bugs.llvm.org/show_bug.cgi?id=40434

llvm-svn: 352048

4 files changed, 200 insertions, 0 deletions

diff --git a/llvm/test/CodeGen/X86/vector-shuffle-256-v16.ll b/llvm/test/CodeGen/X86/vector-shuffle-256-v16.ll
index 461246d80a8..4e447d9429d 100644
--- a/llvm/test/CodeGen/X86/vector-shuffle-256-v16.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-256-v16.ll
@@ -4605,3 +4605,20 @@ define <16 x i16> @insert_dup_elt3_mem_v16i16_i32(i32* %ptr) #0 {
   %tmp3 = shufflevector <8 x i16> %tmp2, <8 x i16> undef, <16 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
   ret <16 x i16> %tmp3
 }
+
+define <16 x i16> @unpckh_v16i16(<16 x i16> %x, <16 x i16> %y) {
+; AVX1-LABEL: unpckh_v16i16:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
+; AVX1-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
+; AVX1-NEXT:    retq
+;
+; AVX2OR512VL-LABEL: unpckh_v16i16:
+; AVX2OR512VL:       # %bb.0:
+; AVX2OR512VL-NEXT:    vextracti128 $1, %ymm1, %xmm1
+; AVX2OR512VL-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
+; AVX2OR512VL-NEXT:    retq
+  %unpckh = shufflevector <16 x i16> %x, <16 x i16> %y, <16 x i32> <i32 4, i32 28, i32 5, i32 29, i32 6, i32 30, i32 7, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  ret <16 x i16> %unpckh
+}
+
diff --git a/llvm/test/CodeGen/X86/vector-shuffle-256-v32.ll b/llvm/test/CodeGen/X86/vector-shuffle-256-v32.ll
index ba0e2086f48..d268364fda8 100644
--- a/llvm/test/CodeGen/X86/vector-shuffle-256-v32.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-256-v32.ll
@@ -3321,3 +3321,20 @@ define <32 x i8> @zeroable_src_to_zext(<32 x i8> %a0) {
   %2 = shufflevector <32 x i8> %1, <32 x i8> <i8 0, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 0, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef>, <32 x i32> <i32 8, i32 9, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 10, i32 11, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 20, i32 21, i32 48, i32 48, i32 48, i32 48, i32 48, i32 48, i32 22, i32 23, i32 48, i32 48, i32 48, i32 48, i32 48, i32 48>
   ret <32 x i8> %2
 }
+
+define <32 x i8> @unpckh_v32i8(<32 x i8> %x, <32 x i8> %y) {
+; AVX1-LABEL: unpckh_v32i8:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
+; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm0 = xmm0[8],xmm1[8],xmm0[9],xmm1[9],xmm0[10],xmm1[10],xmm0[11],xmm1[11],xmm0[12],xmm1[12],xmm0[13],xmm1[13],xmm0[14],xmm1[14],xmm0[15],xmm1[15]
+; AVX1-NEXT:    retq
+;
+; AVX2OR512VL-LABEL: unpckh_v32i8:
+; AVX2OR512VL:       # %bb.0:
+; AVX2OR512VL-NEXT:    vextracti128 $1, %ymm1, %xmm1
+; AVX2OR512VL-NEXT:    vpunpckhbw {{.*#+}} xmm0 = xmm0[8],xmm1[8],xmm0[9],xmm1[9],xmm0[10],xmm1[10],xmm0[11],xmm1[11],xmm0[12],xmm1[12],xmm0[13],xmm1[13],xmm0[14],xmm1[14],xmm0[15],xmm1[15]
+; AVX2OR512VL-NEXT:    retq
+  %unpckh = shufflevector <32 x i8> %x, <32 x i8> %y, <32 x i32> <i32 8, i32 56, i32 9, i32 57, i32 10, i32 58, i32 11, i32 59, i32 12, i32 60, i32 13, i32 61, i32 14, i32 62, i32 15, i32 63, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  ret <32 x i8> %unpckh
+}
+
diff --git a/llvm/test/CodeGen/X86/vector-shuffle-256-v4.ll b/llvm/test/CodeGen/X86/vector-shuffle-256-v4.ll
index b3750b74ad3..83316d0a80c 100644
--- a/llvm/test/CodeGen/X86/vector-shuffle-256-v4.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-256-v4.ll
@@ -1903,3 +1903,64 @@ define <8 x i32> @shuffle_v8i32_0zzzzzzz_optsize(<8 x i32> %a) optsize {
   %b = shufflevector <8 x i32> %a, <8 x i32> zeroinitializer, <8 x i32> <i32 0, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   ret <8 x i32> %b
 }
+
+; FIXME: AVX1 lowering is better than AVX2 (and AVX512?)
+
+define <4 x i64> @unpckh_v4i64(<4 x i64> %x, <4 x i64> %y) {
+; AVX1-LABEL: unpckh_v4i64:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
+; AVX1-NEXT:    vunpckhpd {{.*#+}} xmm0 = xmm0[1],xmm1[1]
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: unpckh_v4i64:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vunpckhpd {{.*#+}} ymm0 = ymm0[1],ymm1[1],ymm0[3],ymm1[3]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
+; AVX2-NEXT:    retq
+;
+; AVX512VL-SLOW-LABEL: unpckh_v4i64:
+; AVX512VL-SLOW:       # %bb.0:
+; AVX512VL-SLOW-NEXT:    vunpckhpd {{.*#+}} ymm0 = ymm0[1],ymm1[1],ymm0[3],ymm1[3]
+; AVX512VL-SLOW-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
+; AVX512VL-SLOW-NEXT:    retq
+;
+; AVX512VL-FAST-LABEL: unpckh_v4i64:
+; AVX512VL-FAST:       # %bb.0:
+; AVX512VL-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [1,7,3,7]
+; AVX512VL-FAST-NEXT:    vpermt2q %ymm1, %ymm2, %ymm0
+; AVX512VL-FAST-NEXT:    retq
+  %unpckh = shufflevector <4 x i64> %x, <4 x i64> %y, <4 x i32> <i32 1, i32 7, i32 undef, i32 undef>
+  ret <4 x i64> %unpckh
+}
+
+; FIXME: AVX1 lowering is better than AVX2 (and AVX512?)
+
+define <4 x double> @unpckh_v4f64(<4 x double> %x, <4 x double> %y) {
+; AVX1-LABEL: unpckh_v4f64:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
+; AVX1-NEXT:    vunpckhpd {{.*#+}} xmm0 = xmm0[1],xmm1[1]
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: unpckh_v4f64:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vunpckhpd {{.*#+}} ymm0 = ymm0[1],ymm1[1],ymm0[3],ymm1[3]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
+; AVX2-NEXT:    retq
+;
+; AVX512VL-SLOW-LABEL: unpckh_v4f64:
+; AVX512VL-SLOW:       # %bb.0:
+; AVX512VL-SLOW-NEXT:    vunpckhpd {{.*#+}} ymm0 = ymm0[1],ymm1[1],ymm0[3],ymm1[3]
+; AVX512VL-SLOW-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
+; AVX512VL-SLOW-NEXT:    retq
+;
+; AVX512VL-FAST-LABEL: unpckh_v4f64:
+; AVX512VL-FAST:       # %bb.0:
+; AVX512VL-FAST-NEXT:    vmovapd {{.*#+}} ymm2 = [1,7,3,7]
+; AVX512VL-FAST-NEXT:    vpermt2pd %ymm1, %ymm2, %ymm0
+; AVX512VL-FAST-NEXT:    retq
+  %unpckh = shufflevector <4 x double> %x, <4 x double> %y, <4 x i32> <i32 1, i32 7, i32 undef, i32 undef>
+  ret <4 x double> %unpckh
+}
+
diff --git a/llvm/test/CodeGen/X86/vector-shuffle-256-v8.ll b/llvm/test/CodeGen/X86/vector-shuffle-256-v8.ll
index addf2d2563f..b2187397bb1 100644
--- a/llvm/test/CodeGen/X86/vector-shuffle-256-v8.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-256-v8.ll
@@ -2877,3 +2877,108 @@ entry:
   %tmp6 = insertelement <8 x float> %tmp5, float %z, i32 5
   ret <8 x float> %tmp6
 }
+
+; FIXME: AVX1 lowering is better than AVX2 (and AVX512?)
+; PR40434: https://bugs.llvm.org/show_bug.cgi?id=40434
+
+define <8 x i32> @unpckh_v8i32(<8 x i32> %x, <8 x i32> %y) {
+; AVX1-LABEL: unpckh_v8i32:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
+; AVX1-NEXT:    vunpckhps {{.*#+}} xmm0 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: unpckh_v8i32:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
+; AVX2-NEXT:    vmovaps {{.*#+}} ymm1 = <2,6,3,7,u,u,u,u>
+; AVX2-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-NEXT:    retq
+;
+; AVX512VL-LABEL: unpckh_v8i32:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm2 = <2,14,3,15,u,u,u,u>
+; AVX512VL-NEXT:    vpermt2d %ymm1, %ymm2, %ymm0
+; AVX512VL-NEXT:    retq
+  %unpckh = shufflevector <8 x i32> %x, <8 x i32> %y, <8 x i32> <i32 2, i32 14, i32 3, i32 15, i32 undef, i32 undef, i32 undef, i32 undef>
+  ret <8 x i32> %unpckh
+}
+
+; FIXME: Same as above but with floats. AVX1 lowering is better than AVX2 (and AVX512?)
+
+define <8 x float> @unpckh_v8f32(<8 x float> %x, <8 x float> %y) {
+; AVX1-LABEL: unpckh_v8f32:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
+; AVX1-NEXT:    vunpckhps {{.*#+}} xmm0 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: unpckh_v8f32:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
+; AVX2-NEXT:    vmovaps {{.*#+}} ymm1 = <2,6,3,7,u,u,u,u>
+; AVX2-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-NEXT:    retq
+;
+; AVX512VL-LABEL: unpckh_v8f32:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vmovaps {{.*#+}} ymm2 = <2,14,3,15,u,u,u,u>
+; AVX512VL-NEXT:    vpermt2ps %ymm1, %ymm2, %ymm0
+; AVX512VL-NEXT:    retq
+  %unpckh = shufflevector <8 x float> %x, <8 x float> %y, <8 x i32> <i32 2, i32 14, i32 3, i32 15, i32 undef, i32 undef, i32 undef, i32 undef>
+  ret <8 x float> %unpckh
+}
+
+; FIXME: Why are integer and FP (below) lowering different for AVX1?
+
+define <8 x i32> @lowhalf_v8i32(<8 x i32> %x, <8 x i32> %y) {
+; AVX1-LABEL: lowhalf_v8i32:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
+; AVX1-NEXT:    vpermilps {{.*#+}} xmm1 = xmm1[0,2,2,2]
+; AVX1-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[2,2,3,3]
+; AVX1-NEXT:    vblendps {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: lowhalf_v8i32:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
+; AVX2-NEXT:    vmovaps {{.*#+}} ymm1 = <2,6,3,6,u,u,u,u>
+; AVX2-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-NEXT:    retq
+;
+; AVX512VL-LABEL: lowhalf_v8i32:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm2 = <2,14,3,14,u,u,u,u>
+; AVX512VL-NEXT:    vpermt2d %ymm1, %ymm2, %ymm0
+; AVX512VL-NEXT:    retq
+  %r = shufflevector <8 x i32> %x, <8 x i32> %y, <8 x i32> <i32 2, i32 14, i32 3, i32 14, i32 undef, i32 undef, i32 undef, i32 undef>
+  ret <8 x i32> %r
+}
+
+; FIXME: AVX1 lowering is better than AVX2 (and AVX512?)
+
+define <8 x float> @lowhalf_v8f32(<8 x float> %x, <8 x float> %y) {
+; AVX1-LABEL: lowhalf_v8f32:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
+; AVX1-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[2,3],xmm1[2,2]
+; AVX1-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[0,2,1,3]
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: lowhalf_v8f32:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
+; AVX2-NEXT:    vmovaps {{.*#+}} ymm1 = <2,6,3,6,u,u,u,u>
+; AVX2-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-NEXT:    retq
+;
+; AVX512VL-LABEL: lowhalf_v8f32:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vmovaps {{.*#+}} ymm2 = <2,14,3,14,u,u,u,u>
+; AVX512VL-NEXT:    vpermt2ps %ymm1, %ymm2, %ymm0
+; AVX512VL-NEXT:    retq
+  %r = shufflevector <8 x float> %x, <8 x float> %y, <8 x i32> <i32 2, i32 14, i32 3, i32 14, i32 undef, i32 undef, i32 undef, i32 undef>
+  ret <8 x float> %r
+}
+