From bd2d90f5a53e66b4f515c5d2876afa8a97b78ca6 Mon Sep 17 00:00:00 2001
From: Bruno Cardoso Lopes <bruno.cardoso@gmail.com>
Date: Fri, 6 Aug 2010 20:03:27 +0000
Subject: [PATCH] Patterns to match AVX 256-bit permutation intrinsics

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@110468 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/X86/X86InstrFragmentsSIMD.td |  2 ++
 lib/Target/X86/X86InstrSSE.td           | 59 +++++++++++++++++++++++++--------
 2 files changed, 48 insertions(+), 13 deletions(-)

diff --git a/lib/Target/X86/X86InstrFragmentsSIMD.td b/lib/Target/X86/X86InstrFragmentsSIMD.td
index 8a1245c5994..a00b0803a3e 100644
--- a/lib/Target/X86/X86InstrFragmentsSIMD.td
+++ b/lib/Target/X86/X86InstrFragmentsSIMD.td
@@ -216,6 +216,8 @@ def memopv16i8 : PatFrag<(ops node:$ptr), (v16i8 (memop node:$ptr))>;
 def memopv32i8 : PatFrag<(ops node:$ptr), (v32i8 (memop node:$ptr))>;
 def memopv8f32 : PatFrag<(ops node:$ptr), (v8f32 (memop node:$ptr))>;
 def memopv4f64 : PatFrag<(ops node:$ptr), (v4f64 (memop node:$ptr))>;
+def memopv4i64 : PatFrag<(ops node:$ptr), (v4i64 (memop node:$ptr))>;
+def memopv8i32 : PatFrag<(ops node:$ptr), (v8i32 (memop node:$ptr))>;
 
 // SSSE3 uses MMX registers for some instructions. They aren't aligned on a
 // 16-byte boundary.
diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index 594d7f615dd..135850fe126 100644
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -5359,29 +5359,44 @@ defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd">;
 
 // Permute Floating-Point Values
 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
-                      RegisterClass RC, X86MemOperand x86memop> {
+                      RegisterClass RC, X86MemOperand x86memop_f,
+                      X86MemOperand x86memop_i, PatFrag f_frag, PatFrag i_frag,
+                      Intrinsic IntVar, Intrinsic IntImm> {
   def rr  : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
              (ins RC:$src1, RC:$src2),
              !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
-             []>, VEX_4V;
+             [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
   def rm  : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
-             (ins RC:$src1, x86memop:$src2),
+             (ins RC:$src1, x86memop_i:$src2),
              !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
-             []>, VEX_4V;
+             [(set RC:$dst, (IntVar RC:$src1, (i_frag addr:$src2)))]>, VEX_4V;
+
   def ri  : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
              (ins RC:$src1, i8imm:$src2),
              !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
-             []>, VEX;
+             [(set RC:$dst, (IntImm RC:$src1, imm:$src2))]>, VEX;
   def mi  : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
-             (ins x86memop:$src1, i8imm:$src2),
+             (ins x86memop_f:$src1, i8imm:$src2),
              !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
-             []>, VEX;
-}
-
-defm VPERMILPS  : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem>;
-defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem>;
-defm VPERMILPD  : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem>;
-defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem>;
+             [(set RC:$dst, (IntImm (f_frag addr:$src1), imm:$src2))]>, VEX;
+}
+
+defm VPERMILPS  : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
+                             memopv4f32, memopv4i32,
+                             int_x86_avx_vpermilvar_ps,
+                             int_x86_avx_vpermil_ps>;
+defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
+                             memopv8f32, memopv8i32,
+                             int_x86_avx_vpermilvar_ps_256,
+                             int_x86_avx_vpermil_ps_256>;
+defm VPERMILPD  : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
+                             memopv2f64, memopv2i64,
+                             int_x86_avx_vpermilvar_pd,
+                             int_x86_avx_vpermil_pd>;
+defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
+                             memopv4f64, memopv4i64,
+                             int_x86_avx_vpermilvar_pd_256,
+                             int_x86_avx_vpermil_pd_256>;
 
 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
           (ins VR256:$src1, VR256:$src2, i8imm:$src3),
@@ -5401,3 +5416,21 @@ def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper", []>, VEX,
                 Requires<[HasAVX]>;
 
 } // isAsmParserOnly
+
+def : Pat<(int_x86_avx_vperm2f128_ps_256 VR256:$src1, VR256:$src2, imm:$src3),
+          (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
+def : Pat<(int_x86_avx_vperm2f128_pd_256 VR256:$src1, VR256:$src2, imm:$src3),
+          (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
+def : Pat<(int_x86_avx_vperm2f128_si_256 VR256:$src1, VR256:$src2, imm:$src3),
+          (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
+
+def : Pat<(int_x86_avx_vperm2f128_ps_256
+                  VR256:$src1, (memopv8f32 addr:$src2), imm:$src3),
+          (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
+def : Pat<(int_x86_avx_vperm2f128_pd_256
+                  VR256:$src1, (memopv4f64 addr:$src2), imm:$src3),
+          (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
+def : Pat<(int_x86_avx_vperm2f128_si_256
+                  VR256:$src1, (memopv8i32 addr:$src2), imm:$src3),
+          (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
+
-- 
2.11.0