[x86] add tests for shift-trunc-shift; NFC

More coverage for a possible generic transform.

2 files changed, 99 insertions, 0 deletions

diff --git a/llvm/test/CodeGen/X86/shift-amount-mod.ll b/llvm/test/CodeGen/X86/shift-amount-mod.ll
index 70fa32027c7..bccb3607c98 100644
--- a/llvm/test/CodeGen/X86/shift-amount-mod.ll
+++ b/llvm/test/CodeGen/X86/shift-amount-mod.ll
@@ -1552,3 +1552,25 @@ define i64 @reg64_lshr_by_masked_negated_unfolded_add_b(i64 %val, i64 %a, i64 %b
   %shifted = lshr i64 %val, %negaaddbitwidthaddb
   ret i64 %shifted
 }
+
+define i16 @sh_trunc_sh(i64 %x) {
+; X32-LABEL: sh_trunc_sh:
+; X32:       # %bb.0:
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    shrl $4, %eax
+; X32-NEXT:    andl $15, %eax
+; X32-NEXT:    # kill: def $ax killed $ax killed $eax
+; X32-NEXT:    retl
+;
+; X64-LABEL: sh_trunc_sh:
+; X64:       # %bb.0:
+; X64-NEXT:    shrq $24, %rdi
+; X64-NEXT:    movzwl %di, %eax
+; X64-NEXT:    shrl $12, %eax
+; X64-NEXT:    # kill: def $ax killed $ax killed $eax
+; X64-NEXT:    retq
+  %s = lshr i64 %x, 24
+  %t = trunc i64 %s to i16
+  %r = lshr i16 %t, 12
+  ret i16 %r
+}
diff --git a/llvm/test/CodeGen/X86/vector-shift-lshr-256.ll b/llvm/test/CodeGen/X86/vector-shift-lshr-256.ll
index 698a45fad4d..c448921db7d 100644
--- a/llvm/test/CodeGen/X86/vector-shift-lshr-256.ll
+++ b/llvm/test/CodeGen/X86/vector-shift-lshr-256.ll
@@ -1394,3 +1394,80 @@ define <32 x i8> @splatconstant_shift_v32i8(<32 x i8> %a) nounwind {
   %shift = lshr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
   ret <32 x i8> %shift
 }
+
+define <4 x i32> @sh_trunc_sh_vec(<4 x i64> %x) {
+; AVX1-LABEL: sh_trunc_sh_vec:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; AVX1-NEXT:    vpsrlq $24, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlq $24, %xmm0, %xmm0
+; AVX1-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
+; AVX1-NEXT:    vpsrld $12, %xmm0, %xmm0
+; AVX1-NEXT:    vzeroupper
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: sh_trunc_sh_vec:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpsrlq $24, %ymm0, %ymm0
+; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; AVX2-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
+; AVX2-NEXT:    vpsrld $12, %xmm0, %xmm0
+; AVX2-NEXT:    vzeroupper
+; AVX2-NEXT:    retq
+;
+; XOPAVX1-LABEL: sh_trunc_sh_vec:
+; XOPAVX1:       # %bb.0:
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; XOPAVX1-NEXT:    vpperm {{.*#+}} xmm0 = xmm0[3,4,5,6,11,12,13,14],xmm1[3,4,5,6,11,12,13,14]
+; XOPAVX1-NEXT:    vpsrld $12, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vzeroupper
+; XOPAVX1-NEXT:    retq
+;
+; XOPAVX2-LABEL: sh_trunc_sh_vec:
+; XOPAVX2:       # %bb.0:
+; XOPAVX2-NEXT:    vpsrlq $24, %ymm0, %ymm0
+; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; XOPAVX2-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
+; XOPAVX2-NEXT:    vpsrld $12, %xmm0, %xmm0
+; XOPAVX2-NEXT:    vzeroupper
+; XOPAVX2-NEXT:    retq
+;
+; AVX512-LABEL: sh_trunc_sh_vec:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpsrlq $24, %ymm0, %ymm0
+; AVX512-NEXT:    vpmovqd %zmm0, %ymm0
+; AVX512-NEXT:    vpsrld $12, %xmm0, %xmm0
+; AVX512-NEXT:    vzeroupper
+; AVX512-NEXT:    retq
+;
+; AVX512VL-LABEL: sh_trunc_sh_vec:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vpsrlq $24, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpmovqd %ymm0, %xmm0
+; AVX512VL-NEXT:    vpsrld $12, %xmm0, %xmm0
+; AVX512VL-NEXT:    vzeroupper
+; AVX512VL-NEXT:    retq
+;
+; X32-AVX1-LABEL: sh_trunc_sh_vec:
+; X32-AVX1:       # %bb.0:
+; X32-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; X32-AVX1-NEXT:    vpsrlq $24, %xmm1, %xmm1
+; X32-AVX1-NEXT:    vpsrlq $24, %xmm0, %xmm0
+; X32-AVX1-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
+; X32-AVX1-NEXT:    vpsrld $12, %xmm0, %xmm0
+; X32-AVX1-NEXT:    vzeroupper
+; X32-AVX1-NEXT:    retl
+;
+; X32-AVX2-LABEL: sh_trunc_sh_vec:
+; X32-AVX2:       # %bb.0:
+; X32-AVX2-NEXT:    vpsrlq $24, %ymm0, %ymm0
+; X32-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; X32-AVX2-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
+; X32-AVX2-NEXT:    vpsrld $12, %xmm0, %xmm0
+; X32-AVX2-NEXT:    vzeroupper
+; X32-AVX2-NEXT:    retl
+  %s = lshr <4 x i64> %x, <i64 24, i64 24, i64 24, i64 24>
+  %t = trunc <4 x i64> %s to <4 x i32>
+  %r = lshr <4 x i32> %t, <i32 12, i32 12, i32 12, i32 12>
+  ret <4 x i32> %r
+}