More AVX2 instructions and their intrinsics.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@143895 91177308-0d34-0410-b5e6-96231b3b80d8
diff --git a/lib/Target/X86/Disassembler/X86DisassemblerDecoder.c b/lib/Target/X86/Disassembler/X86DisassemblerDecoder.c
index f9b0fe5..1a24807 100644
--- a/lib/Target/X86/Disassembler/X86DisassemblerDecoder.c
+++ b/lib/Target/X86/Disassembler/X86DisassemblerDecoder.c
@@ -773,17 +773,20 @@
if (insn->rexPrefix & 0x08)
attrMask |= ATTR_REXW;
-
+
if (getIDWithAttrMask(&instructionID, insn, attrMask))
return -1;
-
+
/* The following clauses compensate for limitations of the tables. */
-
- if ((attrMask & ATTR_VEXL) && (attrMask & ATTR_REXW)) {
+
+ if ((attrMask & ATTR_VEXL) && (attrMask & ATTR_REXW) &&
+ !(attrMask & ATTR_OPSIZE)) {
/*
* Some VEX instructions ignore the L-bit, but use the W-bit. Normally L-bit
* has precedence since there are no L-bit with W-bit entries in the tables.
* So if the L-bit isn't significant we should use the W-bit instead.
+ * We only need to do this if the instruction doesn't specify OpSize since
+ * there is a VEX_L_W_OPSIZE table.
*/
const struct InstructionSpecifier *spec;
diff --git a/lib/Target/X86/Disassembler/X86DisassemblerDecoderCommon.h b/lib/Target/X86/Disassembler/X86DisassemblerDecoderCommon.h
index 8b79335..a7ef0cc 100644
--- a/lib/Target/X86/Disassembler/X86DisassemblerDecoderCommon.h
+++ b/lib/Target/X86/Disassembler/X86DisassemblerDecoderCommon.h
@@ -111,7 +111,8 @@
ENUM_ENTRY(IC_VEX_L, 3, "requires VEX and the L prefix") \
ENUM_ENTRY(IC_VEX_L_XS, 4, "requires VEX and the L and XS prefix")\
ENUM_ENTRY(IC_VEX_L_XD, 4, "requires VEX and the L and XD prefix")\
- ENUM_ENTRY(IC_VEX_L_OPSIZE, 4, "requires VEX, L, and OpSize")
+ ENUM_ENTRY(IC_VEX_L_OPSIZE, 4, "requires VEX, L, and OpSize") \
+ ENUM_ENTRY(IC_VEX_L_W_OPSIZE, 5, "requires VEX, L, W and OpSize")
#define ENUM_ENTRY(n, r, d) n,
diff --git a/lib/Target/X86/X86InstrFormats.td b/lib/Target/X86/X86InstrFormats.td
index b7c172e..ecd6a93 100644
--- a/lib/Target/X86/X86InstrFormats.td
+++ b/lib/Target/X86/X86InstrFormats.td
@@ -460,7 +460,7 @@
list<dag> pattern>
: I<o, F, outs, ins, asm, pattern, SSEPackedInt>, T8, OpSize,
Requires<[HasAVX2]>;
-class AVX2Ii8<bits<8> o, Format F, dag outs, dag ins, string asm,
+class AVX2AIi8<bits<8> o, Format F, dag outs, dag ins, string asm,
list<dag> pattern>
: Ii8<o, F, outs, ins, asm, pattern, SSEPackedInt>, TA, OpSize,
Requires<[HasAVX2]>;
diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index acd9a80..de7326a 100644
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -7310,14 +7310,17 @@
//===----------------------------------------------------------------------===//
// VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
//
+let neverHasSideEffects = 1 in {
def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
(ins VR256:$src1, VR256:$src2, i8imm:$src3),
"vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
[]>, VEX_4V;
+let mayLoad = 1 in
def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
(ins VR256:$src1, f256mem:$src2, i8imm:$src3),
"vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
[]>, VEX_4V;
+}
def : Pat<(int_x86_avx_vperm2f128_ps_256 VR256:$src1, VR256:$src2, imm:$src3),
(VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
@@ -7402,18 +7405,18 @@
// AVX2 Instructions
//===----------------------------------------------------------------------===//
-/// AVX2I_binop_rmi_int - AVX2 binary operator with 8-bit immediate
-multiclass AVX2I_binop_rmi_int<bits<8> opc, string OpcodeStr,
+/// AVX2_binop_rmi_int - AVX2 binary operator with 8-bit immediate
+multiclass AVX2_binop_rmi_int<bits<8> opc, string OpcodeStr,
Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
X86MemOperand x86memop> {
let isCommutable = 1 in
- def rri : AVX2Ii8<opc, MRMSrcReg, (outs RC:$dst),
+ def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
(ins RC:$src1, RC:$src2, u32u8imm:$src3),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
[(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
VEX_4V;
- def rmi : AVX2Ii8<opc, MRMSrcMem, (outs RC:$dst),
+ def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
(ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
@@ -7424,10 +7427,10 @@
}
let isCommutable = 0 in {
-defm VPBLENDD : AVX2I_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
- VR128, memopv16i8, i128mem>;
-defm VPBLENDDY : AVX2I_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
- VR256, memopv32i8, i256mem>;
+defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
+ VR128, memopv16i8, i128mem>;
+defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
+ VR256, memopv32i8, i256mem>;
}
//===----------------------------------------------------------------------===//
@@ -7465,3 +7468,62 @@
defm VPBROADCASTQ : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
int_x86_avx2_pbroadcastq_128,
int_x86_avx2_pbroadcastq_256>;
+
+//===----------------------------------------------------------------------===//
+// VPERM - Permute instructions
+//
+
+multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
+ Intrinsic Int> {
+ def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
+ (ins VR256:$src1, VR256:$src2),
+ !strconcat(OpcodeStr,
+ "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ [(set VR256:$dst, (Int VR256:$src1, VR256:$src2))]>, VEX_4V;
+ def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
+ (ins VR256:$src1, i256mem:$src2),
+ !strconcat(OpcodeStr,
+ "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ [(set VR256:$dst, (Int VR256:$src1, (mem_frag addr:$src2)))]>,
+ VEX_4V;
+}
+
+defm VPERMD : avx2_perm<0x36, "vpermd", memopv8i32, int_x86_avx2_permd>;
+defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, int_x86_avx2_permps>;
+
+multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
+ Intrinsic Int> {
+ def Yrr : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
+ (ins VR256:$src1, i8imm:$src2),
+ !strconcat(OpcodeStr,
+ "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ [(set VR256:$dst, (Int VR256:$src1, imm:$src2))]>, VEX;
+ def Yrm : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
+ (ins i256mem:$src1, i8imm:$src2),
+ !strconcat(OpcodeStr,
+ "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ [(set VR256:$dst, (Int (mem_frag addr:$src1), imm:$src2))]>,
+ VEX;
+}
+
+defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, int_x86_avx2_permq>,
+ VEX_W;
+defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, int_x86_avx2_permpd>,
+ VEX_W;
+
+//===----------------------------------------------------------------------===//
+// VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
+//
+def VPERM2I128rr : AVXAIi8<0x46, MRMSrcReg, (outs VR256:$dst),
+ (ins VR256:$src1, VR256:$src2, i8imm:$src3),
+ "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
+ [(set VR256:$dst,
+ (int_x86_avx2_vperm2i128 VR256:$src1, VR256:$src2, imm:$src3))]>,
+ VEX_4V;
+def VPERM2I128rm : AVXAIi8<0x46, MRMSrcMem, (outs VR256:$dst),
+ (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
+ "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
+ [(set VR256:$dst,
+ (int_x86_avx2_vperm2i128 VR256:$src1, (memopv4i64 addr:$src2),
+ imm:$src3))]>,
+ VEX_4V;