Some instructions were missing, other implemented falsely. this patch aims at amendin...
authorCoby Tayree <coby.tayree@intel.com>
Sun, 20 Nov 2016 17:09:56 +0000 (17:09 +0000)
committerCoby Tayree <coby.tayree@intel.com>
Sun, 20 Nov 2016 17:09:56 +0000 (17:09 +0000)
vcvtps2pd
vcvtudq2pd
vcvtps2qq
vcvttps2qq
vcvtps2uqq
vcvttps2uqq

variants are:

[Dst]XMM(zero-masked/merge-masked/unmasked)
[Src]Mem64

Differential Revision: https://reviews.llvm.org/D26799

llvm-svn: 287500

llvm/lib/Target/X86/X86InstrAVX512.td
llvm/test/MC/X86/intel-syntax-x86-64-avx512f_vl.s
llvm/test/MC/X86/intel-syntax-x86-avx512dq_vl.s [new file with mode: 0644]

index b1b7b27..6a4bd48 100644 (file)
@@ -6096,14 +6096,14 @@ def : Pat<(f32 (fpround FR64X:$src)),
 multiclass avx512_vcvt_fp<bits<8> opc, string OpcodeStr, X86VectorVTInfo _,
                          X86VectorVTInfo _Src, SDNode OpNode,
                          string Broadcast = _.BroadcastStr,
-                         string Alias = ""> {
+                         string Alias = "", X86MemOperand MemOp = _Src.MemOp> {
 
   defm rr : AVX512_maskable<opc, MRMSrcReg, _, (outs _.RC:$dst),
                          (ins _Src.RC:$src), OpcodeStr, "$src", "$src",
                          (_.VT (OpNode (_Src.VT _Src.RC:$src)))>, EVEX;
 
   defm rm : AVX512_maskable<opc, MRMSrcMem, _, (outs _.RC:$dst),
-                         (ins _Src.MemOp:$src), OpcodeStr#Alias, "$src", "$src",
+                         (ins MemOp:$src), OpcodeStr#Alias, "$src", "$src",
                          (_.VT (OpNode (_Src.VT
                              (bitconvert (_Src.LdFrag addr:$src)))))>, EVEX;
 
@@ -6144,7 +6144,7 @@ multiclass avx512_cvtps2pd<bits<8> opc, string OpcodeStr> {
   }
   let Predicates = [HasVLX] in {
     defm Z128 : avx512_vcvt_fp<opc, OpcodeStr, v2f64x_info, v4f32x_info,
-                               X86vfpext, "{1to2}">, EVEX_V128;
+                               X86vfpext, "{1to2}", "", f64mem>, EVEX_V128;
     defm Z256 : avx512_vcvt_fp<opc, OpcodeStr, v4f64x_info, v4f32x_info, fpextend>,
                                      EVEX_V256;
   }
@@ -6203,7 +6203,7 @@ multiclass avx512_cvtdq2pd<bits<8> opc, string OpcodeStr, SDNode OpNode,
 
   let Predicates = [HasVLX] in {
     defm Z128 : avx512_vcvt_fp<opc, OpcodeStr, v2f64x_info, v4i32x_info,
-                                     OpNode128, "{1to2}">, EVEX_V128;
+                                     OpNode128, "{1to2}", "", i64mem>, EVEX_V128;
     defm Z256 : avx512_vcvt_fp<opc, OpcodeStr, v4f64x_info, v4i32x_info, OpNode>,
                                      EVEX_V256;
   }
@@ -6375,7 +6375,7 @@ multiclass avx512_cvtps2qq<bits<8> opc, string OpcodeStr,
     // Explicitly specified broadcast string, since we take only 2 elements
     // from v4f32x_info source
     defm Z128 : avx512_vcvt_fp<opc, OpcodeStr, v2i64x_info, v4f32x_info, OpNode,
-                               "{1to2}">, EVEX_V128;
+                               "{1to2}", "", f64mem>, EVEX_V128;
     defm Z256 : avx512_vcvt_fp<opc, OpcodeStr, v4i64x_info, v4f32x_info, OpNode>,
                                EVEX_V256;
   }
@@ -6393,7 +6393,7 @@ multiclass avx512_cvttps2qq<bits<8> opc, string OpcodeStr,
     // Explicitly specified broadcast string, since we take only 2 elements
     // from v4f32x_info source
     defm Z128 : avx512_vcvt_fp<opc, OpcodeStr, v2i64x_info, v4f32x_info, OpNode,
-                               "{1to2}">, EVEX_V128;
+                               "{1to2}", "", f64mem>, EVEX_V128;
     defm Z256 : avx512_vcvt_fp<opc, OpcodeStr, v4i64x_info, v4f32x_info, OpNode>,
                                EVEX_V256;
   }
@@ -6428,8 +6428,8 @@ multiclass avx512_cvtqq2ps<bits<8> opc, string OpcodeStr,
   }
 }
 
-defm VCVTDQ2PD : avx512_cvtdq2pd<0xE6, "vcvtdq2pd", sint_to_fp, X86cvtdq2pd>, XS,
-                                EVEX_CD8<32, CD8VH>;
+defm VCVTDQ2PD : avx512_cvtdq2pd<0xE6, "vcvtdq2pd", sint_to_fp, X86cvtdq2pd>,
+                                XS, EVEX_CD8<32, CD8VH>;
 
 defm VCVTDQ2PS : avx512_cvtdq2ps<0x5B, "vcvtdq2ps", sint_to_fp,
                                 X86VSintToFpRnd>,
index 4f4c6d1..0e812dd 100644 (file)
 // CHECK: vcvtuqq2ps xmm16, ymmword ptr [rax]
 // CHECK: encoding: [0x62,0xe1,0xff,0x28,0x7a,0x00]
           vcvtuqq2psy xmm16, ymmword ptr [rax]
+
+// CHECK: vcvtps2pd xmm1 {k2} {z}, qword ptr [rcx + 128]
+// CHECK:  encoding: [0x62,0xf1,0x7c,0x8a,0x5a,0x49,0x10]
+          vcvtps2pd xmm1 {k2} {z}, qword ptr [rcx+0x80]
+
+// CHECK: vcvtps2pd xmm1 {k2}, qword ptr [rcx + 128]
+// CHECK:  encoding: [0x62,0xf1,0x7c,0x0a,0x5a,0x49,0x10]
+          vcvtps2pd xmm1 {k2}, qword ptr [rcx+0x80]
+
+// CHECK: vcvtudq2pd xmm2 {k2} {z}, qword ptr [rcx + 128]
+// CHECK: encoding: [0x62,0xf1,0x7e,0x8a,0x7a,0x51,0x10]
+          vcvtudq2pd xmm2 {k2} {z}, qword ptr [rcx+0x80]
+
+// CHECK: vcvtudq2pd xmm2 {k2}, qword ptr [rcx + 128]
+// CHECK: encoding: [0x62,0xf1,0x7e,0x0a,0x7a,0x51,0x10]
+          vcvtudq2pd xmm2 {k2}, qword ptr [rcx+0x80]
+
+// CHECK: vcvtudq2pd xmm2, qword ptr [rcx + 128]
+// CHECK: encoding: [0x62,0xf1,0x7e,0x08,0x7a,0x51,0x10]
+          vcvtudq2pd xmm2, qword ptr [rcx+0x80]
+
+// CHECK: vcvtdq2pd xmm2 {k1}, qword ptr [rcx]
+// CHECK: encoding: [0x62,0xf1,0x7e,0x09,0xe6,0x11]
+          vcvtdq2pd xmm2 {k1}, qword ptr [rcx]
+
+// CHECK: vcvtdq2pd xmm2 {k1} {z}, qword ptr [rcx]
+// CHECK: encoding: [0x62,0xf1,0x7e,0x89,0xe6,0x11]
+          vcvtdq2pd xmm2 {k1} {z}, qword ptr [rcx]
+
diff --git a/llvm/test/MC/X86/intel-syntax-x86-avx512dq_vl.s b/llvm/test/MC/X86/intel-syntax-x86-avx512dq_vl.s
new file mode 100644 (file)
index 0000000..9fa0838
--- /dev/null
@@ -0,0 +1,98 @@
+// RUN: llvm-mc -triple x86_64-unknown-unknown -mcpu=knl -mattr=+avx512vl -mattr=+avx512dq -x86-asm-syntax=intel -output-asm-variant=1 --show-encoding %s | FileCheck %s
+
+// CHECK:  vcvtps2qq xmm2 {k2} {z}, qword ptr [rcx + 128]
+// CHECK:  encoding: [0x62,0xf1,0x7d,0x8a,0x7b,0x51,0x10]
+          vcvtps2qq xmm2 {k2} {z}, qword ptr [rcx + 0x80]
+
+// CHECK:  vcvtps2qq xmm2 {k2}, qword ptr [rcx + 128]
+// CHECK:  encoding: [0x62,0xf1,0x7d,0x0a,0x7b,0x51,0x10]
+          vcvtps2qq xmm2 {k2}, qword ptr [rcx + 0x80]
+
+// CHECK:  vcvtps2qq xmm2, qword ptr [rcx + 128]
+// CHECK:  encoding: [0x62,0xf1,0x7d,0x08,0x7b,0x51,0x10]
+          vcvtps2qq xmm2, qword ptr [rcx + 0x80]
+
+// CHECK:  vcvttps2qq xmm1 {k2} {z}, qword ptr [rcx + 128]
+// CHECK:  encoding: [0x62,0xf1,0x7d,0x8a,0x7a,0x49,0x10]
+          vcvttps2qq xmm1 {k2} {z}, qword ptr [rcx + 0x80]
+
+// CHECK:  vcvttps2qq xmm1 {k2}, qword ptr [rcx + 128]
+// CHECK:  encoding: [0x62,0xf1,0x7d,0x0a,0x7a,0x49,0x10]
+          vcvttps2qq xmm1 {k2}, qword ptr [rcx + 0x80]
+
+// CHECK:  vcvttps2qq xmm1, qword ptr [rcx + 128]
+// CHECK:  encoding: [0x62,0xf1,0x7d,0x08,0x7a,0x49,0x10]
+          vcvttps2qq xmm1, qword ptr [rcx + 0x80]
+
+// CHECK:  vcvtps2uqq xmm1 {k2} {z}, qword ptr [rcx + 128]
+// CHECK:  encoding: [0x62,0xf1,0x7d,0x8a,0x79,0x49,0x10]
+          vcvtps2uqq xmm1 {k2} {z}, qword ptr [rcx + 128]
+
+// CHECK:  vcvtps2uqq xmm1 {k2}, qword ptr [rcx + 128]
+// CHECK:  encoding: [0x62,0xf1,0x7d,0x0a,0x79,0x49,0x10]
+          vcvtps2uqq xmm1 {k2}, qword ptr [rcx + 128]
+
+// CHECK:  vcvtps2uqq xmm1, qword ptr [rcx + 128]
+// CHECK:  encoding: [0x62,0xf1,0x7d,0x08,0x79,0x49,0x10]
+          vcvtps2uqq xmm1, qword ptr [rcx + 128]
+
+// CHECK:  vcvttps2uqq xmm1 {k2} {z}, qword ptr [rcx + 128]
+// CHECK:  encoding: [0x62,0xf1,0x7d,0x8a,0x78,0x49,0x10]
+          vcvttps2uqq xmm1 {k2} {z}, qword ptr [rcx + 128]
+
+// CHECK:  vcvttps2uqq xmm1 {k2}, qword ptr [rcx + 128]
+// CHECK:  encoding: [0x62,0xf1,0x7d,0x0a,0x78,0x49,0x10]
+          vcvttps2uqq xmm1 {k2}, qword ptr [rcx + 128]
+
+// CHECK:  vcvttps2uqq xmm1, qword ptr [rcx + 128]
+// CHECK:  encoding: [0x62,0xf1,0x7d,0x08,0x78,0x49,0x10]
+          vcvttps2uqq xmm1, qword ptr [rcx + 128]
+// RUN: llvm-mc -triple x86_64-unknown-unknown -mcpu=knl -mattr=+avx512vl -mattr=+avx512dq -x86-asm-syntax=intel -output-asm-variant=1 --show-encoding %s | FileCheck %s
+
+// CHECK:  vcvtps2qq xmm2 {k2} {z}, qword ptr [rcx + 128]
+// CHECK:  encoding: [0x62,0xf1,0x7d,0x8a,0x7b,0x51,0x10]
+          vcvtps2qq xmm2 {k2} {z}, qword ptr [rcx + 0x80]
+
+// CHECK:  vcvtps2qq xmm2 {k2}, qword ptr [rcx + 128]
+// CHECK:  encoding: [0x62,0xf1,0x7d,0x0a,0x7b,0x51,0x10]
+          vcvtps2qq xmm2 {k2}, qword ptr [rcx + 0x80]
+
+// CHECK:  vcvtps2qq xmm2, qword ptr [rcx + 128]
+// CHECK:  encoding: [0x62,0xf1,0x7d,0x08,0x7b,0x51,0x10]
+          vcvtps2qq xmm2, qword ptr [rcx + 0x80]
+
+// CHECK:  vcvttps2qq xmm1 {k2} {z}, qword ptr [rcx + 128]
+// CHECK:  encoding: [0x62,0xf1,0x7d,0x8a,0x7a,0x49,0x10]
+          vcvttps2qq xmm1 {k2} {z}, qword ptr [rcx + 0x80]
+
+// CHECK:  vcvttps2qq xmm1 {k2}, qword ptr [rcx + 128]
+// CHECK:  encoding: [0x62,0xf1,0x7d,0x0a,0x7a,0x49,0x10]
+          vcvttps2qq xmm1 {k2}, qword ptr [rcx + 0x80]
+
+// CHECK:  vcvttps2qq xmm1, qword ptr [rcx + 128]
+// CHECK:  encoding: [0x62,0xf1,0x7d,0x08,0x7a,0x49,0x10]
+          vcvttps2qq xmm1, qword ptr [rcx + 0x80]
+
+// CHECK:  vcvtps2uqq xmm1 {k2} {z}, qword ptr [rcx + 128]
+// CHECK:  encoding: [0x62,0xf1,0x7d,0x8a,0x79,0x49,0x10]
+          vcvtps2uqq xmm1 {k2} {z}, qword ptr [rcx + 128]
+
+// CHECK:  vcvtps2uqq xmm1 {k2}, qword ptr [rcx + 128]
+// CHECK:  encoding: [0x62,0xf1,0x7d,0x0a,0x79,0x49,0x10]
+          vcvtps2uqq xmm1 {k2}, qword ptr [rcx + 128]
+
+// CHECK:  vcvtps2uqq xmm1, qword ptr [rcx + 128]
+// CHECK:  encoding: [0x62,0xf1,0x7d,0x08,0x79,0x49,0x10]
+          vcvtps2uqq xmm1, qword ptr [rcx + 128]
+
+// CHECK:  vcvttps2uqq xmm1 {k2} {z}, qword ptr [rcx + 128]
+// CHECK:  encoding: [0x62,0xf1,0x7d,0x8a,0x78,0x49,0x10]
+          vcvttps2uqq xmm1 {k2} {z}, qword ptr [rcx + 128]
+
+// CHECK:  vcvttps2uqq xmm1 {k2}, qword ptr [rcx + 128]
+// CHECK:  encoding: [0x62,0xf1,0x7d,0x0a,0x78,0x49,0x10]
+          vcvttps2uqq xmm1 {k2}, qword ptr [rcx + 128]
+
+// CHECK:  vcvttps2uqq xmm1, qword ptr [rcx + 128]
+// CHECK:  encoding: [0x62,0xf1,0x7d,0x08,0x78,0x49,0x10]
+          vcvttps2uqq xmm1, qword ptr [rcx + 128]