Introduce 'UseSSEx' to force SSE legacy encoding

- Add 'UseSSEx' to force SSE legacy insn not being selected when AVX is
  enabled.

  As the penalty of inter-mixing SSE and AVX instructions, we need
  prevent SSE legacy insn from being generated except explicitly
  specified through some intrinsics. For patterns supported by both
  SSE and AVX, so far, we force AVX insn will be tried first relying on
  AddedComplexity or position in td file. It's error-prone and
  introduces bugs accidentally.

  'UseSSEx' is disabled when AVX is turned on. For SSE insns inherited
  by AVX, we need this predicate to force VEX encoding or SSE legacy
  encoding only.

  For insns not inherited by AVX, we still use the previous predicates,
  i.e. 'HasSSEx'. So far, these insns fall into the following
  categories:
  * SSE insns with MMX operands
  * SSE insns with GPR/MEM operands only (xFENCE, PREFETCH, CLFLUSH,
    CRC, and etc.)
  * SSE4A insns.
  * MMX insns.
  * x87 insns added by SSE.

2 test cases are modified:

 - test/CodeGen/X86/fast-isel-x86-64.ll
   AVX code generation is different from SSE one. 'vcvtsi2sdq' cannot be
   selected by fast-isel due to complicated pattern and fast-isel
   fallback to materialize it from constant pool.

 - test/CodeGen/X86/widen_load-1.ll
   AVX code generation is different from SSE one after fixing SSE/AVX
   inter-mixing. Exec-domain fixing prefers 'vmovapd' instead of
   'vmovaps'.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@162919 91177308-0d34-0410-b5e6-96231b3b80d8

1 files changed, 33 insertions, 32 deletions

diff --git a/lib/Target/X86/X86InstrMMX.td b/lib/Target/X86/X86InstrMMX.td
index c8f40bbb49..bd5485840d 100644
--- a/lib/Target/X86/X86InstrMMX.td
+++ b/lib/Target/X86/X86InstrMMX.td
@@ -118,11 +118,11 @@ let Constraints = "$src1 = $dst" in {
 /// Unary MMX instructions requiring SSSE3.
 multiclass SS3I_unop_rm_int_mm<bits<8> opc, string OpcodeStr,
                                Intrinsic IntId64, OpndItins itins> {
-  def rr64 : SS38I<opc, MRMSrcReg, (outs VR64:$dst), (ins VR64:$src),
+  def rr64 : MMXSS38I<opc, MRMSrcReg, (outs VR64:$dst), (ins VR64:$src),
                    !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
                    [(set VR64:$dst, (IntId64 VR64:$src))], itins.rr>;
 
-  def rm64 : SS38I<opc, MRMSrcMem, (outs VR64:$dst), (ins i64mem:$src),
+  def rm64 : MMXSS38I<opc, MRMSrcMem, (outs VR64:$dst), (ins i64mem:$src),
                    !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
                    [(set VR64:$dst,
                      (IntId64 (bitconvert (memopmmx addr:$src))))],
@@ -134,11 +134,11 @@ let ImmT = NoImm, Constraints = "$src1 = $dst" in {
 multiclass SS3I_binop_rm_int_mm<bits<8> opc, string OpcodeStr,
                              Intrinsic IntId64, OpndItins itins> {
   let isCommutable = 0 in
-  def rr64 : SS38I<opc, MRMSrcReg, (outs VR64:$dst),
+  def rr64 : MMXSS38I<opc, MRMSrcReg, (outs VR64:$dst),
        (ins VR64:$src1, VR64:$src2),
         !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
        [(set VR64:$dst, (IntId64 VR64:$src1, VR64:$src2))], itins.rr>;
-  def rm64 : SS38I<opc, MRMSrcMem, (outs VR64:$dst),
+  def rm64 : MMXSS38I<opc, MRMSrcMem, (outs VR64:$dst),
        (ins VR64:$src1, i64mem:$src2),
         !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
        [(set VR64:$dst,
@@ -149,11 +149,11 @@ multiclass SS3I_binop_rm_int_mm<bits<8> opc, string OpcodeStr,
 
 /// PALIGN MMX instructions (require SSSE3).
 multiclass ssse3_palign_mm<string asm, Intrinsic IntId> {
-  def R64irr  : SS3AI<0x0F, MRMSrcReg, (outs VR64:$dst),
+  def R64irr  : MMXSS3AI<0x0F, MRMSrcReg, (outs VR64:$dst),
       (ins VR64:$src1, VR64:$src2, i8imm:$src3),
       !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"), 
       [(set VR64:$dst, (IntId VR64:$src1, VR64:$src2, (i8 imm:$src3)))]>;
-  def R64irm  : SS3AI<0x0F, MRMSrcMem, (outs VR64:$dst),
+  def R64irm  : MMXSS3AI<0x0F, MRMSrcMem, (outs VR64:$dst),
       (ins VR64:$src1, i64mem:$src2, i8imm:$src3),
       !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
       [(set VR64:$dst, (IntId VR64:$src1,
@@ -163,12 +163,10 @@ multiclass ssse3_palign_mm<string asm, Intrinsic IntId> {
 multiclass sse12_cvt_pint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
                          Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
                          string asm, OpndItins itins, Domain d> {
-  def irr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
-                        [(set DstRC:$dst, (Int SrcRC:$src))], 
-                        itins.rr, d>;
-  def irm : PI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
-                        [(set DstRC:$dst, (Int (ld_frag addr:$src)))], 
-                        itins.rm, d>;
+  def irr : MMXPI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
+                  [(set DstRC:$dst, (Int SrcRC:$src))], itins.rr, d>;
+  def irm : MMXPI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
+                  [(set DstRC:$dst, (Int (ld_frag addr:$src)))], itins.rm, d>;
 }
 
 multiclass sse12_cvt_pint_3addr<bits<8> opc, RegisterClass SrcRC,
@@ -243,29 +241,30 @@ def MMX_MOVQ64mr : MMXI<0x7F, MRMDestMem, (outs), (ins i64mem:$dst, VR64:$src),
                         [(store (x86mmx VR64:$src), addr:$dst)],
                         IIC_MMX_MOVQ_RM>;
 
-def MMX_MOVDQ2Qrr : SDIi8<0xD6, MRMSrcReg, (outs VR64:$dst),
-                          (ins VR128:$src), "movdq2q\t{$src, $dst|$dst, $src}",
-                          [(set VR64:$dst,
-                            (x86mmx (bitconvert
-                            (i64 (vector_extract (v2i64 VR128:$src),
-                                  (iPTR 0))))))],
-                          IIC_MMX_MOVQ_RR>;
-
-def MMX_MOVQ2DQrr : S2SIi8<0xD6, MRMSrcReg, (outs VR128:$dst),
-                            (ins VR64:$src), "movq2dq\t{$src, $dst|$dst, $src}",
-          [(set VR128:$dst,
-            (v2i64 (scalar_to_vector
-                              (i64 (bitconvert (x86mmx VR64:$src))))))],
-                           IIC_MMX_MOVQ_RR>;
+def MMX_MOVDQ2Qrr : MMXSDIi8<0xD6, MRMSrcReg, (outs VR64:$dst),
+                             (ins VR128:$src), "movdq2q\t{$src, $dst|$dst, $src}",
+                             [(set VR64:$dst,
+                               (x86mmx (bitconvert
+                               (i64 (vector_extract (v2i64 VR128:$src),
+                                     (iPTR 0))))))],
+                             IIC_MMX_MOVQ_RR>;
+
+def MMX_MOVQ2DQrr : MMXS2SIi8<0xD6, MRMSrcReg, (outs VR128:$dst),
+                              (ins VR64:$src), "movq2dq\t{$src, $dst|$dst, $src}",
+                              [(set VR128:$dst,
+                                (v2i64
+                                  (scalar_to_vector
+                                    (i64 (bitconvert (x86mmx VR64:$src))))))],
+                              IIC_MMX_MOVQ_RR>;
 
 let neverHasSideEffects = 1 in
-def MMX_MOVQ2FR64rr: S2SIi8<0xD6, MRMSrcReg, (outs FR64:$dst),
-                       (ins VR64:$src), "movq2dq\t{$src, $dst|$dst, $src}", [],
-                       IIC_MMX_MOVQ_RR>;
+def MMX_MOVQ2FR64rr: MMXS2SIi8<0xD6, MRMSrcReg, (outs FR64:$dst),
+                               (ins VR64:$src), "movq2dq\t{$src, $dst|$dst, $src}",
+                               [], IIC_MMX_MOVQ_RR>;
 
-def MMX_MOVFR642Qrr: SDIi8<0xD6, MRMSrcReg, (outs VR64:$dst),
-                       (ins FR64:$src), "movdq2q\t{$src, $dst|$dst, $src}", [],
-                       IIC_MMX_MOVQ_RR>;
+def MMX_MOVFR642Qrr: MMXSDIi8<0xD6, MRMSrcReg, (outs VR64:$dst),
+                              (ins FR64:$src), "movdq2q\t{$src, $dst|$dst, $src}",
+                              [], IIC_MMX_MOVQ_RR>;
 
 def MMX_MOVNTQmr  : MMXI<0xE7, MRMDestMem, (outs), (ins i64mem:$dst, VR64:$src),
                          "movntq\t{$src, $dst|$dst, $src}",
@@ -577,6 +576,7 @@ def MMX_MASKMOVQ64: MMXI64<0xF7, MRMSrcReg, (outs), (ins VR64:$src, VR64:$mask),
                            IIC_MMX_MASKMOV>;
 
 // 64-bit bit convert.
+let Predicates = [HasSSE2] in {
 def : Pat<(x86mmx (bitconvert (i64 GR64:$src))),
           (MMX_MOVD64to64rr GR64:$src)>;
 def : Pat<(i64 (bitconvert (x86mmx VR64:$src))),
@@ -585,5 +585,6 @@ def : Pat<(f64 (bitconvert (x86mmx VR64:$src))),
           (MMX_MOVQ2FR64rr VR64:$src)>;
 def : Pat<(x86mmx (bitconvert (f64 FR64:$src))),
           (MMX_MOVFR642Qrr FR64:$src)>;
+}