[SelectionDAG] Add basic demanded elements support to ComputeNumSignBits for BITCAST nodes

Only adds support to the existing 'large element' scalar/vector to 'small element' vector bitcasts.

The next step would be to support cases where the large elements aren't all sign bits, and determine the small element equivalent based on the demanded elements.

llvm-svn: 340143

2 files changed, 36 insertions, 43 deletions

diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
index f93a2af..7b28a06d 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
@@ -3245,7 +3245,14 @@ unsigned SelectionDAG::ComputeNumSignBits(SDValue Op, const APInt &DemandedElts,
     // Requires handling of DemandedElts and Endianness.
     if ((SrcBits % VTBits) == 0) {
       assert(Op.getValueType().isVector() && "Expected bitcast to vector");
-      Tmp = ComputeNumSignBits(N0, Depth + 1);
+
+      unsigned Scale = SrcBits / VTBits;
+      APInt SrcDemandedElts(NumElts / Scale, 0);
+      for (unsigned i = 0; i != NumElts; ++i)
+        if (DemandedElts[i])
+          SrcDemandedElts.setBit(i / Scale);
+
+      Tmp = ComputeNumSignBits(N0, SrcDemandedElts, Depth + 1);
       if (Tmp == SrcBits)
         return VTBits;
     }
diff --git a/llvm/test/CodeGen/X86/packss.ll b/llvm/test/CodeGen/X86/packss.ll
index 3723ab4..a062365 100644
--- a/llvm/test/CodeGen/X86/packss.ll
+++ b/llvm/test/CodeGen/X86/packss.ll
@@ -107,29 +107,25 @@ define <8 x i16> @trunc_ashr_v4i32_icmp_v4i32(<4 x i32> %a, <4 x i32> %b) nounwi
 define <8 x i16> @trunc_ashr_v4i64_demandedelts(<4 x i64> %a0) {
 ; X86-SSE-LABEL: trunc_ashr_v4i64_demandedelts:
 ; X86-SSE:       # %bb.0:
-; X86-SSE-NEXT:    movdqa %xmm0, %xmm2
+; X86-SSE-NEXT:    movdqa %xmm1, %xmm2
 ; X86-SSE-NEXT:    psllq $63, %xmm2
-; X86-SSE-NEXT:    movdqa %xmm0, %xmm3
+; X86-SSE-NEXT:    movdqa %xmm1, %xmm3
 ; X86-SSE-NEXT:    movsd {{.*#+}} xmm3 = xmm2[0],xmm3[1]
-; X86-SSE-NEXT:    movdqa %xmm1, %xmm2
+; X86-SSE-NEXT:    movdqa %xmm0, %xmm2
 ; X86-SSE-NEXT:    psllq $63, %xmm2
-; X86-SSE-NEXT:    movdqa %xmm1, %xmm4
+; X86-SSE-NEXT:    movdqa %xmm0, %xmm4
 ; X86-SSE-NEXT:    movsd {{.*#+}} xmm4 = xmm2[0],xmm4[1]
 ; X86-SSE-NEXT:    psrlq $63, %xmm4
-; X86-SSE-NEXT:    movsd {{.*#+}} xmm1 = xmm4[0],xmm1[1]
+; X86-SSE-NEXT:    movsd {{.*#+}} xmm0 = xmm4[0],xmm0[1]
 ; X86-SSE-NEXT:    movapd {{.*#+}} xmm2 = [4.940656e-324,-0.000000e+00]
-; X86-SSE-NEXT:    xorpd %xmm2, %xmm1
-; X86-SSE-NEXT:    psubq %xmm2, %xmm1
-; X86-SSE-NEXT:    psrlq $63, %xmm3
-; X86-SSE-NEXT:    movsd {{.*#+}} xmm0 = xmm3[0],xmm0[1]
 ; X86-SSE-NEXT:    xorpd %xmm2, %xmm0
 ; X86-SSE-NEXT:    psubq %xmm2, %xmm0
-; X86-SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
+; X86-SSE-NEXT:    psrlq $63, %xmm3
+; X86-SSE-NEXT:    movsd {{.*#+}} xmm1 = xmm3[0],xmm1[1]
+; X86-SSE-NEXT:    xorpd %xmm2, %xmm1
+; X86-SSE-NEXT:    psubq %xmm2, %xmm1
 ; X86-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
-; X86-SSE-NEXT:    pslld $16, %xmm1
-; X86-SSE-NEXT:    psrad $16, %xmm1
-; X86-SSE-NEXT:    pslld $16, %xmm0
-; X86-SSE-NEXT:    psrad $16, %xmm0
+; X86-SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
 ; X86-SSE-NEXT:    packssdw %xmm1, %xmm0
 ; X86-SSE-NEXT:    retl
 ;
@@ -151,10 +147,7 @@ define <8 x i16> @trunc_ashr_v4i64_demandedelts(<4 x i64> %a0) {
 ; X86-AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
 ; X86-AVX1-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[0,0,0,0,4,4,4,4]
 ; X86-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; X86-AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
-; X86-AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; X86-AVX1-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; X86-AVX1-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; X86-AVX1-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vzeroupper
 ; X86-AVX1-NEXT:    retl
 ;
@@ -167,37 +160,33 @@ define <8 x i16> @trunc_ashr_v4i64_demandedelts(<4 x i64> %a0) {
 ; X86-AVX2-NEXT:    vpxor %ymm2, %ymm0, %ymm0
 ; X86-AVX2-NEXT:    vpsrlvq %ymm1, %ymm0, %ymm0
 ; X86-AVX2-NEXT:    vpsubq %ymm3, %ymm0, %ymm0
-; X86-AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,0,1,0,1,0,1,0,1,0,1,0,1,2,3,16,17,16,17,16,17,16,17,16,17,16,17,16,17,18,19]
-; X86-AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; X86-AVX2-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
+; X86-AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,0,0,0,4,4,4,4]
+; X86-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; X86-AVX2-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vzeroupper
 ; X86-AVX2-NEXT:    retl
 ;
 ; X64-SSE-LABEL: trunc_ashr_v4i64_demandedelts:
 ; X64-SSE:       # %bb.0:
-; X64-SSE-NEXT:    movdqa %xmm0, %xmm2
+; X64-SSE-NEXT:    movdqa %xmm1, %xmm2
 ; X64-SSE-NEXT:    psllq $63, %xmm2
-; X64-SSE-NEXT:    movdqa %xmm0, %xmm3
+; X64-SSE-NEXT:    movdqa %xmm1, %xmm3
 ; X64-SSE-NEXT:    movsd {{.*#+}} xmm3 = xmm2[0],xmm3[1]
-; X64-SSE-NEXT:    movdqa %xmm1, %xmm2
+; X64-SSE-NEXT:    movdqa %xmm0, %xmm2
 ; X64-SSE-NEXT:    psllq $63, %xmm2
-; X64-SSE-NEXT:    movdqa %xmm1, %xmm4
+; X64-SSE-NEXT:    movdqa %xmm0, %xmm4
 ; X64-SSE-NEXT:    movsd {{.*#+}} xmm4 = xmm2[0],xmm4[1]
 ; X64-SSE-NEXT:    psrlq $63, %xmm4
-; X64-SSE-NEXT:    movsd {{.*#+}} xmm1 = xmm4[0],xmm1[1]
+; X64-SSE-NEXT:    movsd {{.*#+}} xmm0 = xmm4[0],xmm0[1]
 ; X64-SSE-NEXT:    movapd {{.*#+}} xmm2 = [1,9223372036854775808]
-; X64-SSE-NEXT:    xorpd %xmm2, %xmm1
-; X64-SSE-NEXT:    psubq %xmm2, %xmm1
-; X64-SSE-NEXT:    psrlq $63, %xmm3
-; X64-SSE-NEXT:    movsd {{.*#+}} xmm0 = xmm3[0],xmm0[1]
 ; X64-SSE-NEXT:    xorpd %xmm2, %xmm0
 ; X64-SSE-NEXT:    psubq %xmm2, %xmm0
-; X64-SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
+; X64-SSE-NEXT:    psrlq $63, %xmm3
+; X64-SSE-NEXT:    movsd {{.*#+}} xmm1 = xmm3[0],xmm1[1]
+; X64-SSE-NEXT:    xorpd %xmm2, %xmm1
+; X64-SSE-NEXT:    psubq %xmm2, %xmm1
 ; X64-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
-; X64-SSE-NEXT:    pslld $16, %xmm1
-; X64-SSE-NEXT:    psrad $16, %xmm1
-; X64-SSE-NEXT:    pslld $16, %xmm0
-; X64-SSE-NEXT:    psrad $16, %xmm0
+; X64-SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
 ; X64-SSE-NEXT:    packssdw %xmm1, %xmm0
 ; X64-SSE-NEXT:    retq
 ;
@@ -220,10 +209,7 @@ define <8 x i16> @trunc_ashr_v4i64_demandedelts(<4 x i64> %a0) {
 ; X64-AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
 ; X64-AVX1-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[0,0,0,0,4,4,4,4]
 ; X64-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; X64-AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
-; X64-AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; X64-AVX1-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; X64-AVX1-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; X64-AVX1-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vzeroupper
 ; X64-AVX1-NEXT:    retq
 ;
@@ -234,9 +220,9 @@ define <8 x i16> @trunc_ashr_v4i64_demandedelts(<4 x i64> %a0) {
 ; X64-AVX2-NEXT:    # ymm1 = mem[0,1,0,1]
 ; X64-AVX2-NEXT:    vpxor %ymm1, %ymm0, %ymm0
 ; X64-AVX2-NEXT:    vpsubq %ymm1, %ymm0, %ymm0
-; X64-AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,0,1,0,1,0,1,0,1,0,1,0,1,2,3,16,17,16,17,16,17,16,17,16,17,16,17,16,17,18,19]
-; X64-AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; X64-AVX2-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
+; X64-AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,0,0,0,4,4,4,4]
+; X64-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; X64-AVX2-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vzeroupper
 ; X64-AVX2-NEXT:    retq
   %1 = shl <4 x i64> %a0, <i64 63, i64 0, i64 63, i64 0>