[X86][SSE] Split ComputeNumSignBits SEXT/AND/XOR/OR demandedelts test

Max depth was being exceeded which could prevent some combines working

llvm-svn: 316871

1 files changed, 65 insertions, 17 deletions

diff --git a/llvm/test/CodeGen/X86/known-signbits-vector.ll b/llvm/test/CodeGen/X86/known-signbits-vector.ll
index 6ff20b5eba8..f88637904a7 100644
--- a/llvm/test/CodeGen/X86/known-signbits-vector.ll
+++ b/llvm/test/CodeGen/X86/known-signbits-vector.ll
@@ -276,8 +276,8 @@ define <2 x double> @signbits_ashr_concat_ashr_extract_sitofp(<2 x i64> %a0, <4
   ret <2 x double> %6
 }
 
-define float @signbits_ashr_sextinreg_bitops_extract_sitofp(<2 x i64> %a0, <2 x i64> %a1, i32 %a2) nounwind {
-; X32-LABEL: signbits_ashr_sextinreg_bitops_extract_sitofp:
+define float @signbits_ashr_sext_sextinreg_and_extract_sitofp(<2 x i64> %a0, <2 x i64> %a1, i32 %a2) nounwind {
+; X32-LABEL: signbits_ashr_sext_sextinreg_and_extract_sitofp:
 ; X32:       # BB#0:
 ; X32-NEXT:    pushl %ebp
 ; X32-NEXT:    movl %esp, %ebp
@@ -300,9 +300,7 @@ define float @signbits_ashr_sextinreg_bitops_extract_sitofp(<2 x i64> %a0, <2 x
 ; X32-NEXT:    vpsrad $20, %xmm1, %xmm2
 ; X32-NEXT:    vpsrlq $20, %xmm1, %xmm1
 ; X32-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
-; X32-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; X32-NEXT:    vpor %xmm1, %xmm2, %xmm1
-; X32-NEXT:    vpxor %xmm0, %xmm1, %xmm0
+; X32-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; X32-NEXT:    vmovq %xmm0, {{[0-9]+}}(%esp)
 ; X32-NEXT:    fildll {{[0-9]+}}(%esp)
 ; X32-NEXT:    fstps {{[0-9]+}}(%esp)
@@ -311,7 +309,7 @@ define float @signbits_ashr_sextinreg_bitops_extract_sitofp(<2 x i64> %a0, <2 x
 ; X32-NEXT:    popl %ebp
 ; X32-NEXT:    retl
 ;
-; X64-LABEL: signbits_ashr_sextinreg_bitops_extract_sitofp:
+; X64-LABEL: signbits_ashr_sext_sextinreg_and_extract_sitofp:
 ; X64:       # BB#0:
 ; X64-NEXT:    vpsrlq $60, %xmm0, %xmm2
 ; X64-NEXT:    vpsrlq $61, %xmm0, %xmm0
@@ -325,21 +323,71 @@ define float @signbits_ashr_sextinreg_bitops_extract_sitofp(<2 x i64> %a0, <2 x
 ; X64-NEXT:    vpsrad $20, %xmm1, %xmm2
 ; X64-NEXT:    vpsrlq $20, %xmm1, %xmm1
 ; X64-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
+; X64-NEXT:    vpand %xmm1, %xmm0, %xmm0
+; X64-NEXT:    vmovq %xmm0, %rax
+; X64-NEXT:    vcvtsi2ssq %rax, %xmm3, %xmm0
+; X64-NEXT:    retq
+  %1 = ashr <2 x i64> %a0, <i64 61, i64 60>
+  %2 = sext i32 %a2 to i64
+  %3 = insertelement <2 x i64> %a1, i64 %2, i32 0
+  %4 = shl <2 x i64> %3, <i64 20, i64 20>
+  %5 = ashr <2 x i64> %4, <i64 20, i64 20>
+  %6 = and <2 x i64> %1, %5
+  %7 = extractelement <2 x i64> %6, i32 0
+  %8 = sitofp i64 %7 to float
+  ret float %8
+}
+
+define float @signbits_ashr_sextvecinreg_bitops_extract_sitofp(<2 x i64> %a0, <4 x i32> %a1) nounwind {
+; X32-LABEL: signbits_ashr_sextvecinreg_bitops_extract_sitofp:
+; X32:       # BB#0:
+; X32-NEXT:    pushl %ebp
+; X32-NEXT:    movl %esp, %ebp
+; X32-NEXT:    andl $-8, %esp
+; X32-NEXT:    subl $16, %esp
+; X32-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,2147483648,0,2147483648]
+; X32-NEXT:    vpsrlq $60, %xmm2, %xmm3
+; X32-NEXT:    vpsrlq $61, %xmm2, %xmm2
+; X32-NEXT:    vpblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm3[4,5,6,7]
+; X32-NEXT:    vpsrlq $60, %xmm0, %xmm3
+; X32-NEXT:    vpsrlq $61, %xmm0, %xmm0
+; X32-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm3[4,5,6,7]
+; X32-NEXT:    vpxor %xmm2, %xmm0, %xmm0
+; X32-NEXT:    vpsubq %xmm2, %xmm0, %xmm0
+; X32-NEXT:    vpmovsxdq %xmm1, %xmm1
+; X32-NEXT:    vpand %xmm1, %xmm0, %xmm2
+; X32-NEXT:    vpor %xmm1, %xmm2, %xmm1
+; X32-NEXT:    vpxor %xmm0, %xmm1, %xmm0
+; X32-NEXT:    vmovq %xmm0, {{[0-9]+}}(%esp)
+; X32-NEXT:    fildll {{[0-9]+}}(%esp)
+; X32-NEXT:    fstps {{[0-9]+}}(%esp)
+; X32-NEXT:    flds {{[0-9]+}}(%esp)
+; X32-NEXT:    movl %ebp, %esp
+; X32-NEXT:    popl %ebp
+; X32-NEXT:    retl
+;
+; X64-LABEL: signbits_ashr_sextvecinreg_bitops_extract_sitofp:
+; X64:       # BB#0:
+; X64-NEXT:    vpsrlq $60, %xmm0, %xmm2
+; X64-NEXT:    vpsrlq $61, %xmm0, %xmm0
+; X64-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm2[4,5,6,7]
+; X64-NEXT:    vmovdqa {{.*#+}} xmm2 = [4,8]
+; X64-NEXT:    vpxor %xmm2, %xmm0, %xmm0
+; X64-NEXT:    vpsubq %xmm2, %xmm0, %xmm0
+; X64-NEXT:    vpmovsxdq %xmm1, %xmm1
 ; X64-NEXT:    vpand %xmm1, %xmm0, %xmm2
 ; X64-NEXT:    vpor %xmm1, %xmm2, %xmm1
 ; X64-NEXT:    vpxor %xmm0, %xmm1, %xmm0
 ; X64-NEXT:    vmovq %xmm0, %rax
 ; X64-NEXT:    vcvtsi2ssq %rax, %xmm3, %xmm0
 ; X64-NEXT:    retq
-  %1  = ashr <2 x i64> %a0, <i64 61, i64 60>
-  %2 = sext i32 %a2 to i64
-  %3  = insertelement <2 x i64> %a1, i64 %2, i32 0
-  %4  = shl <2 x i64> %3, <i64 20, i64 20>
-  %5  = ashr <2 x i64> %4, <i64 20, i64 20>
-  %6  = and <2 x i64> %1, %5
-  %7  = or <2 x i64> %6, %5
-  %8  = xor <2 x i64> %7, %1
-  %9  = extractelement <2 x i64> %8, i32 0
-  %10 = sitofp i64 %9 to float
-  ret float %10
+  %1 = ashr <2 x i64> %a0, <i64 61, i64 60>
+  %2 = shufflevector <4 x i32> %a1, <4 x i32> undef, <2 x i32> <i32 0, i32 1>
+  %3 = sext <2 x i32> %2 to <2 x i64>
+  %4 = and <2 x i64> %1, %3
+  %5 = or <2 x i64> %4, %3
+  %6 = xor <2 x i64> %5, %1
+  %7 = extractelement <2 x i64> %6, i32 0
+  %8 = sitofp i64 %7 to float
+  ret float %8
 }