Add new AVX vpermilps, vpermilpd and vperm2f128 instructions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108984 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index 32730855734d19359736f277f56d55dc758fe6a1..76bc0ff48254793f19a823314c9e04b48edc236a 100644 (file)
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -5041,4 +5041,39 @@ multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr> {
 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps">;
 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> {
+  def rr  : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
+             (ins RC:$src1, RC:$src2),
+             !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+             []>, VEX_4V;
+  def rm  : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
+             (ins RC:$src1, x86memop:$src2),
+             !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+             []>, VEX_4V;
+  def ri  : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
+             (ins RC:$src1, i8imm:$src2),
+             !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+             []>, VEX;
+  def mi  : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
+             (ins x86memop:$src1, i8imm:$src2),
+             !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+             []>, VEX;
+}
+
+defm VPERMILPS  : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem>;
+defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem>;
+defm VPERMILPD  : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem>;
+defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem>;
+
+def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
+          (ins VR256:$src1, VR256:$src2, i8imm:$src3),
+          "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
+          []>, VEX_4V;
+def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
+          (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
+          "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
+          []>, VEX_4V;
+
 } // isAsmParserOnly

diff --git a/test/MC/AsmParser/X86/x86_32-encoding.s b/test/MC/AsmParser/X86/x86_32-encoding.s
index daa8cb583b8f05e050e61edae63d1cd5fcc97dfc..8c62260c3f0be56da93e59af46133b6ab202e518 100644 (file)
--- a/test/MC/AsmParser/X86/x86_32-encoding.s
+++ b/test/MC/AsmParser/X86/x86_32-encoding.s
@@ -13062,3 +13062,75 @@
 // CHECK: encoding: [0xc4,0xe2,0x6d,0x2c,0x28]
           vmaskmovps  (%eax), %ymm2, %ymm5
 
+// CHECK: vpermilps  $7, %xmm1, %xmm5
+// CHECK: encoding: [0xc4,0xe3,0x79,0x04,0xe9,0x07]
+          vpermilps  $7, %xmm1, %xmm5
+
+// CHECK: vpermilps  $7, %ymm5, %ymm1
+// CHECK: encoding: [0xc4,0xe3,0x7d,0x04,0xcd,0x07]
+          vpermilps  $7, %ymm5, %ymm1
+
+// CHECK: vpermilps  $7, (%eax), %xmm5
+// CHECK: encoding: [0xc4,0xe3,0x79,0x04,0x28,0x07]
+          vpermilps  $7, (%eax), %xmm5
+
+// CHECK: vpermilps  $7, (%eax), %ymm5
+// CHECK: encoding: [0xc4,0xe3,0x7d,0x04,0x28,0x07]
+          vpermilps  $7, (%eax), %ymm5
+
+// CHECK: vpermilps  %xmm1, %xmm5, %xmm1
+// CHECK: encoding: [0xc4,0xe2,0x51,0x0c,0xc9]
+          vpermilps  %xmm1, %xmm5, %xmm1
+
+// CHECK: vpermilps  %ymm1, %ymm5, %ymm1
+// CHECK: encoding: [0xc4,0xe2,0x55,0x0c,0xc9]
+          vpermilps  %ymm1, %ymm5, %ymm1
+
+// CHECK: vpermilps  (%eax), %xmm5, %xmm3
+// CHECK: encoding: [0xc4,0xe2,0x51,0x0c,0x18]
+          vpermilps  (%eax), %xmm5, %xmm3
+
+// CHECK: vpermilps  (%eax), %ymm5, %ymm1
+// CHECK: encoding: [0xc4,0xe2,0x55,0x0c,0x08]
+          vpermilps  (%eax), %ymm5, %ymm1
+
+// CHECK: vpermilpd  $7, %xmm1, %xmm5
+// CHECK: encoding: [0xc4,0xe3,0x79,0x05,0xe9,0x07]
+          vpermilpd  $7, %xmm1, %xmm5
+
+// CHECK: vpermilpd  $7, %ymm5, %ymm1
+// CHECK: encoding: [0xc4,0xe3,0x7d,0x05,0xcd,0x07]
+          vpermilpd  $7, %ymm5, %ymm1
+
+// CHECK: vpermilpd  $7, (%eax), %xmm5
+// CHECK: encoding: [0xc4,0xe3,0x79,0x05,0x28,0x07]
+          vpermilpd  $7, (%eax), %xmm5
+
+// CHECK: vpermilpd  $7, (%eax), %ymm5
+// CHECK: encoding: [0xc4,0xe3,0x7d,0x05,0x28,0x07]
+          vpermilpd  $7, (%eax), %ymm5
+
+// CHECK: vpermilpd  %xmm1, %xmm5, %xmm1
+// CHECK: encoding: [0xc4,0xe2,0x51,0x0d,0xc9]
+          vpermilpd  %xmm1, %xmm5, %xmm1
+
+// CHECK: vpermilpd  %ymm1, %ymm5, %ymm1
+// CHECK: encoding: [0xc4,0xe2,0x55,0x0d,0xc9]
+          vpermilpd  %ymm1, %ymm5, %ymm1
+
+// CHECK: vpermilpd  (%eax), %xmm5, %xmm3
+// CHECK: encoding: [0xc4,0xe2,0x51,0x0d,0x18]
+          vpermilpd  (%eax), %xmm5, %xmm3
+
+// CHECK: vpermilpd  (%eax), %ymm5, %ymm1
+// CHECK: encoding: [0xc4,0xe2,0x55,0x0d,0x08]
+          vpermilpd  (%eax), %ymm5, %ymm1
+
+// CHECK: vperm2f128  $7, %ymm2, %ymm5, %ymm1
+// CHECK: encoding: [0xc4,0xe3,0x55,0x06,0xca,0x07]
+          vperm2f128  $7, %ymm2, %ymm5, %ymm1
+
+// CHECK: vperm2f128  $7, (%eax), %ymm5, %ymm1
+// CHECK: encoding: [0xc4,0xe3,0x55,0x06,0x08,0x07]
+          vperm2f128  $7, (%eax), %ymm5, %ymm1
+

diff --git a/test/MC/AsmParser/X86/x86_64-encoding.s b/test/MC/AsmParser/X86/x86_64-encoding.s
index 5affb4e0d69fc73efcb8f491b2aecc3dadaaddee..5593897374bd6db11f6d9344b7cd163178e8bd2f 100644 (file)
--- a/test/MC/AsmParser/X86/x86_64-encoding.s
+++ b/test/MC/AsmParser/X86/x86_64-encoding.s
@@ -3136,3 +3136,75 @@ pshufb   CPI1_0(%rip), %xmm1
 // CHECK: encoding: [0xc4,0x62,0x1d,0x2c,0x10]
           vmaskmovps  (%rax), %ymm12, %ymm10
 
+// CHECK: vpermilps  $7, %xmm11, %xmm10
+// CHECK: encoding: [0xc4,0x43,0x79,0x04,0xd3,0x07]
+          vpermilps  $7, %xmm11, %xmm10
+
+// CHECK: vpermilps  $7, %ymm10, %ymm11
+// CHECK: encoding: [0xc4,0x43,0x7d,0x04,0xda,0x07]
+          vpermilps  $7, %ymm10, %ymm11
+
+// CHECK: vpermilps  $7, (%rax), %xmm10
+// CHECK: encoding: [0xc4,0x63,0x79,0x04,0x10,0x07]
+          vpermilps  $7, (%rax), %xmm10
+
+// CHECK: vpermilps  $7, (%rax), %ymm10
+// CHECK: encoding: [0xc4,0x63,0x7d,0x04,0x10,0x07]
+          vpermilps  $7, (%rax), %ymm10
+
+// CHECK: vpermilps  %xmm11, %xmm10, %xmm11
+// CHECK: encoding: [0xc4,0x42,0x29,0x0c,0xdb]
+          vpermilps  %xmm11, %xmm10, %xmm11
+
+// CHECK: vpermilps  %ymm11, %ymm10, %ymm11
+// CHECK: encoding: [0xc4,0x42,0x2d,0x0c,0xdb]
+          vpermilps  %ymm11, %ymm10, %ymm11
+
+// CHECK: vpermilps  (%rax), %xmm10, %xmm13
+// CHECK: encoding: [0xc4,0x62,0x29,0x0c,0x28]
+          vpermilps  (%rax), %xmm10, %xmm13
+
+// CHECK: vpermilps  (%rax), %ymm10, %ymm11
+// CHECK: encoding: [0xc4,0x62,0x2d,0x0c,0x18]
+          vpermilps  (%rax), %ymm10, %ymm11
+
+// CHECK: vpermilpd  $7, %xmm11, %xmm10
+// CHECK: encoding: [0xc4,0x43,0x79,0x05,0xd3,0x07]
+          vpermilpd  $7, %xmm11, %xmm10
+
+// CHECK: vpermilpd  $7, %ymm10, %ymm11
+// CHECK: encoding: [0xc4,0x43,0x7d,0x05,0xda,0x07]
+          vpermilpd  $7, %ymm10, %ymm11
+
+// CHECK: vpermilpd  $7, (%rax), %xmm10
+// CHECK: encoding: [0xc4,0x63,0x79,0x05,0x10,0x07]
+          vpermilpd  $7, (%rax), %xmm10
+
+// CHECK: vpermilpd  $7, (%rax), %ymm10
+// CHECK: encoding: [0xc4,0x63,0x7d,0x05,0x10,0x07]
+          vpermilpd  $7, (%rax), %ymm10
+
+// CHECK: vpermilpd  %xmm11, %xmm10, %xmm11
+// CHECK: encoding: [0xc4,0x42,0x29,0x0d,0xdb]
+          vpermilpd  %xmm11, %xmm10, %xmm11
+
+// CHECK: vpermilpd  %ymm11, %ymm10, %ymm11
+// CHECK: encoding: [0xc4,0x42,0x2d,0x0d,0xdb]
+          vpermilpd  %ymm11, %ymm10, %ymm11
+
+// CHECK: vpermilpd  (%rax), %xmm10, %xmm13
+// CHECK: encoding: [0xc4,0x62,0x29,0x0d,0x28]
+          vpermilpd  (%rax), %xmm10, %xmm13
+
+// CHECK: vpermilpd  (%rax), %ymm10, %ymm11
+// CHECK: encoding: [0xc4,0x62,0x2d,0x0d,0x18]
+          vpermilpd  (%rax), %ymm10, %ymm11
+
+// CHECK: vperm2f128  $7, %ymm12, %ymm10, %ymm11
+// CHECK: encoding: [0xc4,0x43,0x2d,0x06,0xdc,0x07]
+          vperm2f128  $7, %ymm12, %ymm10, %ymm11
+
+// CHECK: vperm2f128  $7, (%rax), %ymm10, %ymm11
+// CHECK: encoding: [0xc4,0x63,0x2d,0x06,0x18,0x07]
+          vperm2f128  $7, (%rax), %ymm10, %ymm11
+