[X86][MC] Support Enc/Dec for EGPR for promoted CRC32 (#76434)

R16-R31 was added into GPRs in #70958, This patch supports the encoding/decoding for promoted CRC32 instruction in EVEX space. RFC: https://discourse.llvm.org/t/rfc-design-for-apx-feature-egpr-and-ndd-support/73031/4
llvm · Jan 2, 2024 · d8db273 · d8db273
1 parent 7a3b0cb
commit d8db273
Show file tree

Hide file tree

Showing 6 changed files with 294 additions and 4 deletions.
diff --git a/llvm/lib/Target/X86/X86InstrSSE.td b/llvm/lib/Target/X86/X86InstrSSE.td
@@ -6663,18 +6663,18 @@ let Defs = [ECX, EFLAGS], Uses = [EAX, EDX], hasSideEffects = 0 in {
 class Crc32r<X86TypeInfo t, RegisterClass rc, SDPatternOperator node>
   : ITy<0xF1, MRMSrcReg, t, (outs rc:$dst), (ins rc:$src1, t.RegClass:$src2),
       "crc32", binop_args, [(set rc:$dst, (node rc:$src1, t.RegClass:$src2))]>,
-    Sched<[WriteCRC32]> {
+    Sched<[WriteCRC32]>, NoCD8 {
   let Constraints = "$src1 = $dst";
 }
 
 class Crc32m<X86TypeInfo t, RegisterClass rc, SDPatternOperator node>
   : ITy<0xF1, MRMSrcMem, t, (outs rc:$dst), (ins rc:$src1, t.MemOperand:$src2),
       "crc32", binop_args, [(set rc:$dst, (node rc:$src1, (load addr:$src2)))]>,
-    Sched<[WriteCRC32.Folded, WriteCRC32.ReadAfterFold]> {
+    Sched<[WriteCRC32.Folded, WriteCRC32.ReadAfterFold]>, NoCD8 {
   let Constraints = "$src1 = $dst";
 }
 
-let Predicates = [HasCRC32], OpMap = T8, OpPrefix = XD in {
+let Predicates = [HasCRC32, NoEGPR], OpMap = T8, OpPrefix = XD in {
   def CRC32r32r8  : Crc32r<Xi8, GR32, int_x86_sse42_crc32_32_8>;
   def CRC32r32m8  : Crc32m<Xi8, GR32, int_x86_sse42_crc32_32_8>;
   def CRC32r32r16 : Crc32r<Xi16, GR32, int_x86_sse42_crc32_32_16>, OpSize16;
@@ -6688,6 +6688,20 @@ let Predicates = [HasCRC32], OpMap = T8, OpPrefix = XD in {
     def CRC32r64m8 : Crc32m<Xi8, GR64, null_frag>, REX_W;
 }
 
+let Predicates = [HasCRC32, HasEGPR, In64BitMode], OpMap = T_MAP4, OpEnc = EncEVEX in {
+  def CRC32r32r8_EVEX  : Crc32r<Xi8, GR32, int_x86_sse42_crc32_32_8>;
+  def CRC32r32m8_EVEX  : Crc32m<Xi8, GR32, int_x86_sse42_crc32_32_8>;
+  def CRC32r32r16_EVEX : Crc32r<Xi16, GR32, int_x86_sse42_crc32_32_16>, PD;
+  def CRC32r32m16_EVEX : Crc32m<Xi16, GR32, int_x86_sse42_crc32_32_16>, PD;
+  def CRC32r32r32_EVEX : Crc32r<Xi32, GR32, int_x86_sse42_crc32_32_32>;
+  def CRC32r32m32_EVEX : Crc32m<Xi32, GR32, int_x86_sse42_crc32_32_32>;
+  def CRC32r64r64_EVEX : Crc32r<Xi64, GR64, int_x86_sse42_crc32_64_64>;
+  def CRC32r64m64_EVEX : Crc32m<Xi64, GR64, int_x86_sse42_crc32_64_64>;
+  def CRC32r64r8_EVEX : Crc32r<Xi8, GR64, null_frag>, REX_W;
+  let mayLoad = 1 in
+    def CRC32r64m8_EVEX : Crc32m<Xi8, GR64, null_frag>, REX_W;
+}
+
 //===----------------------------------------------------------------------===//
 // SHA-NI Instructions
 //===----------------------------------------------------------------------===//

diff --git a/llvm/test/MC/Disassembler/X86/apx/crc32.txt b/llvm/test/MC/Disassembler/X86/apx/crc32.txt
@@ -0,0 +1,90 @@
+# RUN: llvm-mc --disassemble %s -triple=x86_64 | FileCheck %s --check-prefixes=ATT
+# RUN: llvm-mc --disassemble %s -triple=x86_64 -x86-asm-syntax=intel --output-asm-variant=1 | FileCheck %s --check-prefixes=INTEL
+
+# ATT:   crc32b	%al, %ebx
+# INTEL: crc32	ebx, al
+0x62,0xf4,0x7c,0x08,0xf0,0xd8
+
+# ATT:   crc32b	%al, %rbx
+# INTEL: crc32	rbx, al
+0x62,0xf4,0xfc,0x08,0xf0,0xd8
+
+# ATT:   crc32w	%ax, %ebx
+# INTEL: crc32	ebx, ax
+0x62,0xf4,0x7d,0x08,0xf1,0xd8
+
+# ATT:   crc32l	%eax, %ebx
+# INTEL: crc32	ebx, eax
+0x62,0xf4,0x7c,0x08,0xf1,0xd8
+
+# ATT:   crc32q	%rax, %rbx
+# INTEL: crc32	rbx, rax
+0x62,0xf4,0xfc,0x08,0xf1,0xd8
+
+# ATT:   crc32w	291(%rax,%rbx,4), %ecx
+# INTEL: crc32	ecx, word ptr [rax + 4*rbx + 291]
+0x62,0xf4,0x7d,0x08,0xf1,0x8c,0x98,0x23,0x01,0x00,0x00
+
+# ATT:   crc32l	291(%rax,%rbx,4), %ecx
+# INTEL: crc32	ecx, dword ptr [rax + 4*rbx + 291]
+0x62,0xf4,0x7c,0x08,0xf1,0x8c,0x98,0x23,0x01,0x00,0x00
+
+# ATT:   crc32b	291(%rax,%rbx,4), %rcx
+# INTEL: crc32	rcx, byte ptr [rax + 4*rbx + 291]
+0x62,0xf4,0xfc,0x08,0xf0,0x8c,0x98,0x23,0x01,0x00,0x00
+
+# ATT:   crc32q	291(%rax,%rbx,4), %rcx
+# INTEL: crc32	rcx, qword ptr [rax + 4*rbx + 291]
+0x62,0xf4,0xfc,0x08,0xf1,0x8c,0x98,0x23,0x01,0x00,0x00
+
+# ATT:   crc32b	%r16b, %r22d
+# INTEL: crc32	r22d, r16b
+0x62,0xec,0x7c,0x08,0xf0,0xf0
+
+# ATT:   crc32b	%r16b, %r23
+# INTEL: crc32	r23, r16b
+0x62,0xec,0xfc,0x08,0xf0,0xf8
+
+# ATT:   crc32w	%r17w, %r22d
+# INTEL: crc32	r22d, r17w
+0x62,0xec,0x7d,0x08,0xf1,0xf1
+
+# ATT:   crc32l	%r18d, %r22d
+# INTEL: crc32	r22d, r18d
+0x62,0xec,0x7c,0x08,0xf1,0xf2
+
+# ATT:   crc32q	%r19, %r23
+# INTEL: crc32	r23, r19
+0x62,0xec,0xfc,0x08,0xf1,0xfb
+
+# ATT:   crc32w	291(%r28,%r29,4), %r18d
+# INTEL: crc32	r18d, word ptr [r28 + 4*r29 + 291]
+0x62,0x8c,0x79,0x08,0xf1,0x94,0xac,0x23,0x01,0x00,0x00
+
+# ATT:   crc32l	291(%r28,%r29,4), %r18d
+# INTEL: crc32	r18d, dword ptr [r28 + 4*r29 + 291]
+0x62,0x8c,0x78,0x08,0xf1,0x94,0xac,0x23,0x01,0x00,0x00
+
+# ATT:   crc32b	291(%r28,%r29,4), %r19
+# INTEL: crc32	r19, byte ptr [r28 + 4*r29 + 291]
+0x62,0x8c,0xf8,0x08,0xf0,0x9c,0xac,0x23,0x01,0x00,0x00
+
+# ATT:   crc32q	291(%r28,%r29,4), %r19
+# INTEL: crc32	r19, qword ptr [r28 + 4*r29 + 291]
+0x62,0x8c,0xf8,0x08,0xf1,0x9c,0xac,0x23,0x01,0x00,0x00
+
+# ATT:   crc32w	123(%r28,%r29,4), %r18d
+# INTEL: crc32	r18d, word ptr [r28 + 4*r29 + 123]
+0x62,0x8c,0x79,0x08,0xf1,0x54,0xac,0x7b
+
+# ATT:   crc32l	123(%r28,%r29,4), %r18d
+# INTEL: crc32	r18d, dword ptr [r28 + 4*r29 + 123]
+0x62,0x8c,0x78,0x08,0xf1,0x54,0xac,0x7b
+
+# ATT:   crc32b	123(%r28,%r29,4), %r19
+# INTEL: crc32	r19, byte ptr [r28 + 4*r29 + 123]
+0x62,0x8c,0xf8,0x08,0xf0,0x5c,0xac,0x7b
+
+# ATT:   crc32q	123(%r28,%r29,4), %r19
+# INTEL: crc32	r19, qword ptr [r28 + 4*r29 + 123]
+0x62,0x8c,0xf8,0x08,0xf1,0x5c,0xac,0x7b
diff --git a/llvm/test/MC/X86/apx/crc32-att.s b/llvm/test/MC/X86/apx/crc32-att.s
@@ -0,0 +1,92 @@
+# RUN: llvm-mc -triple x86_64 --show-encoding %s | FileCheck %s
+# RUN: not llvm-mc -triple i386 -show-encoding %s 2>&1 | FileCheck %s --check-prefix=ERROR
+
+# ERROR-COUNT-22: error:
+# ERROR-NOT: error:
+# CHECK: {evex}	crc32b	%al, %ebx
+# CHECK: encoding: [0x62,0xf4,0x7c,0x08,0xf0,0xd8]
+         {evex}	crc32b	%al, %ebx
+
+# CHECK: {evex}	crc32b	%al, %rbx
+# CHECK: encoding: [0x62,0xf4,0xfc,0x08,0xf0,0xd8]
+         {evex}	crc32b	%al, %rbx
+
+# CHECK: {evex}	crc32w	%ax, %ebx
+# CHECK: encoding: [0x62,0xf4,0x7d,0x08,0xf1,0xd8]
+         {evex}	crc32w	%ax, %ebx
+
+# CHECK: {evex}	crc32l	%eax, %ebx
+# CHECK: encoding: [0x62,0xf4,0x7c,0x08,0xf1,0xd8]
+         {evex}	crc32l	%eax, %ebx
+
+# CHECK: {evex}	crc32q	%rax, %rbx
+# CHECK: encoding: [0x62,0xf4,0xfc,0x08,0xf1,0xd8]
+         {evex}	crc32q	%rax, %rbx
+
+# CHECK: {evex}	crc32w	291(%rax,%rbx,4), %ecx
+# CHECK: encoding: [0x62,0xf4,0x7d,0x08,0xf1,0x8c,0x98,0x23,0x01,0x00,0x00]
+         {evex}	crc32w	291(%rax,%rbx,4), %ecx
+
+# CHECK: {evex}	crc32l	291(%rax,%rbx,4), %ecx
+# CHECK: encoding: [0x62,0xf4,0x7c,0x08,0xf1,0x8c,0x98,0x23,0x01,0x00,0x00]
+         {evex}	crc32l	291(%rax,%rbx,4), %ecx
+
+# CHECK: {evex}	crc32b	291(%rax,%rbx,4), %rcx
+# CHECK: encoding: [0x62,0xf4,0xfc,0x08,0xf0,0x8c,0x98,0x23,0x01,0x00,0x00]
+         {evex}	crc32b	291(%rax,%rbx,4), %rcx
+
+# CHECK: {evex}	crc32q	291(%rax,%rbx,4), %rcx
+# CHECK: encoding: [0x62,0xf4,0xfc,0x08,0xf1,0x8c,0x98,0x23,0x01,0x00,0x00]
+         {evex}	crc32q	291(%rax,%rbx,4), %rcx
+
+# CHECK: crc32b	%r16b, %r22d
+# CHECK: encoding: [0x62,0xec,0x7c,0x08,0xf0,0xf0]
+         crc32b	%r16b, %r22d
+
+# CHECK: crc32b	%r16b, %r23
+# CHECK: encoding: [0x62,0xec,0xfc,0x08,0xf0,0xf8]
+         crc32b	%r16b, %r23
+
+# CHECK: crc32w	%r17w, %r22d
+# CHECK: encoding: [0x62,0xec,0x7d,0x08,0xf1,0xf1]
+         crc32w	%r17w, %r22d
+
+# CHECK: crc32l	%r18d, %r22d
+# CHECK: encoding: [0x62,0xec,0x7c,0x08,0xf1,0xf2]
+         crc32l	%r18d, %r22d
+
+# CHECK: crc32q	%r19, %r23
+# CHECK: encoding: [0x62,0xec,0xfc,0x08,0xf1,0xfb]
+         crc32q	%r19, %r23
+
+# CHECK: crc32w	291(%r28,%r29,4), %r18d
+# CHECK: encoding: [0x62,0x8c,0x79,0x08,0xf1,0x94,0xac,0x23,0x01,0x00,0x00]
+         crc32w	291(%r28,%r29,4), %r18d
+
+# CHECK: crc32l	291(%r28,%r29,4), %r18d
+# CHECK: encoding: [0x62,0x8c,0x78,0x08,0xf1,0x94,0xac,0x23,0x01,0x00,0x00]
+         crc32l	291(%r28,%r29,4), %r18d
+
+# CHECK: crc32b	291(%r28,%r29,4), %r19
+# CHECK: encoding: [0x62,0x8c,0xf8,0x08,0xf0,0x9c,0xac,0x23,0x01,0x00,0x00]
+         crc32b	291(%r28,%r29,4), %r19
+
+# CHECK: crc32q	291(%r28,%r29,4), %r19
+# CHECK: encoding: [0x62,0x8c,0xf8,0x08,0xf1,0x9c,0xac,0x23,0x01,0x00,0x00]
+         crc32q	291(%r28,%r29,4), %r19
+
+# CHECK: crc32w	123(%r28,%r29,4), %r18d
+# CHECK: encoding: [0x62,0x8c,0x79,0x08,0xf1,0x54,0xac,0x7b]
+         crc32w	123(%r28,%r29,4), %r18d
+
+# CHECK: crc32l	123(%r28,%r29,4), %r18d
+# CHECK: encoding: [0x62,0x8c,0x78,0x08,0xf1,0x54,0xac,0x7b]
+         crc32l	123(%r28,%r29,4), %r18d
+
+# CHECK: crc32b	123(%r28,%r29,4), %r19
+# CHECK: encoding: [0x62,0x8c,0xf8,0x08,0xf0,0x5c,0xac,0x7b]
+         crc32b	123(%r28,%r29,4), %r19
+
+# CHECK: crc32q	123(%r28,%r29,4), %r19
+# CHECK: encoding: [0x62,0x8c,0xf8,0x08,0xf1,0x5c,0xac,0x7b]
+         crc32q	123(%r28,%r29,4), %r19
diff --git a/llvm/test/MC/X86/apx/crc32-intel.s b/llvm/test/MC/X86/apx/crc32-intel.s
@@ -0,0 +1,89 @@
+# RUN: llvm-mc -triple x86_64 -x86-asm-syntax=intel -output-asm-variant=1 --show-encoding %s | FileCheck %s
+
+# CHECK: {evex}	crc32	ebx, al
+# CHECK: encoding: [0x62,0xf4,0x7c,0x08,0xf0,0xd8]
+         {evex}	crc32	ebx, al
+
+# CHECK: {evex}	crc32	rbx, al
+# CHECK: encoding: [0x62,0xf4,0xfc,0x08,0xf0,0xd8]
+         {evex}	crc32	rbx, al
+
+# CHECK: {evex}	crc32	ebx, ax
+# CHECK: encoding: [0x62,0xf4,0x7d,0x08,0xf1,0xd8]
+         {evex}	crc32	ebx, ax
+
+# CHECK: {evex}	crc32	ebx, eax
+# CHECK: encoding: [0x62,0xf4,0x7c,0x08,0xf1,0xd8]
+         {evex}	crc32	ebx, eax
+
+# CHECK: {evex}	crc32	rbx, rax
+# CHECK: encoding: [0x62,0xf4,0xfc,0x08,0xf1,0xd8]
+         {evex}	crc32	rbx, rax
+
+# CHECK: {evex}	crc32	ecx, word ptr [rax + 4*rbx + 291]
+# CHECK: encoding: [0x62,0xf4,0x7d,0x08,0xf1,0x8c,0x98,0x23,0x01,0x00,0x00]
+         {evex}	crc32	ecx, word ptr [rax + 4*rbx + 291]
+
+# CHECK: {evex}	crc32	ecx, dword ptr [rax + 4*rbx + 291]
+# CHECK: encoding: [0x62,0xf4,0x7c,0x08,0xf1,0x8c,0x98,0x23,0x01,0x00,0x00]
+         {evex}	crc32	ecx, dword ptr [rax + 4*rbx + 291]
+
+# CHECK: {evex}	crc32	rcx, byte ptr [rax + 4*rbx + 291]
+# CHECK: encoding: [0x62,0xf4,0xfc,0x08,0xf0,0x8c,0x98,0x23,0x01,0x00,0x00]
+         {evex}	crc32	rcx, byte ptr [rax + 4*rbx + 291]
+
+# CHECK: {evex}	crc32	rcx, qword ptr [rax + 4*rbx + 291]
+# CHECK: encoding: [0x62,0xf4,0xfc,0x08,0xf1,0x8c,0x98,0x23,0x01,0x00,0x00]
+         {evex}	crc32	rcx, qword ptr [rax + 4*rbx + 291]
+
+# CHECK: crc32	r22d, r16b
+# CHECK: encoding: [0x62,0xec,0x7c,0x08,0xf0,0xf0]
+         crc32	r22d, r16b
+
+# CHECK: crc32	r23, r16b
+# CHECK: encoding: [0x62,0xec,0xfc,0x08,0xf0,0xf8]
+         crc32	r23, r16b
+
+# CHECK: crc32	r22d, r17w
+# CHECK: encoding: [0x62,0xec,0x7d,0x08,0xf1,0xf1]
+         crc32	r22d, r17w
+
+# CHECK: crc32	r22d, r18d
+# CHECK: encoding: [0x62,0xec,0x7c,0x08,0xf1,0xf2]
+         crc32	r22d, r18d
+
+# CHECK: crc32	r23, r19
+# CHECK: encoding: [0x62,0xec,0xfc,0x08,0xf1,0xfb]
+         crc32	r23, r19
+
+# CHECK: crc32	r18d, word ptr [r28 + 4*r29 + 291]
+# CHECK: encoding: [0x62,0x8c,0x79,0x08,0xf1,0x94,0xac,0x23,0x01,0x00,0x00]
+         crc32	r18d, word ptr [r28 + 4*r29 + 291]
+
+# CHECK: crc32	r18d, dword ptr [r28 + 4*r29 + 291]
+# CHECK: encoding: [0x62,0x8c,0x78,0x08,0xf1,0x94,0xac,0x23,0x01,0x00,0x00]
+         crc32	r18d, dword ptr [r28 + 4*r29 + 291]
+
+# CHECK: crc32	r19, byte ptr [r28 + 4*r29 + 291]
+# CHECK: encoding: [0x62,0x8c,0xf8,0x08,0xf0,0x9c,0xac,0x23,0x01,0x00,0x00]
+         crc32	r19, byte ptr [r28 + 4*r29 + 291]
+
+# CHECK: crc32	r19, qword ptr [r28 + 4*r29 + 291]
+# CHECK: encoding: [0x62,0x8c,0xf8,0x08,0xf1,0x9c,0xac,0x23,0x01,0x00,0x00]
+         crc32	r19, qword ptr [r28 + 4*r29 + 291]
+
+# CHECK: crc32	r18d, word ptr [r28 + 4*r29 + 123]
+# CHECK: encoding: [0x62,0x8c,0x79,0x08,0xf1,0x54,0xac,0x7b]
+         crc32	r18d, word ptr [r28 + 4*r29 + 123]
+
+# CHECK: crc32	r18d, dword ptr [r28 + 4*r29 + 123]
+# CHECK: encoding: [0x62,0x8c,0x78,0x08,0xf1,0x54,0xac,0x7b]
+         crc32	r18d, dword ptr [r28 + 4*r29 + 123]
+
+# CHECK: crc32	r19, byte ptr [r28 + 4*r29 + 123]
+# CHECK: encoding: [0x62,0x8c,0xf8,0x08,0xf0,0x5c,0xac,0x7b]
+         crc32	r19, byte ptr [r28 + 4*r29 + 123]
+
+# CHECK: crc32	r19, qword ptr [r28 + 4*r29 + 123]
+# CHECK: encoding: [0x62,0x8c,0xf8,0x08,0xf1,0x5c,0xac,0x7b]
+         crc32	r19, qword ptr [r28 + 4*r29 + 123]
diff --git a/llvm/test/MC/X86/x86_64-asm-match.s b/llvm/test/MC/X86/x86_64-asm-match.s
@@ -29,7 +29,7 @@
 // CHECK:   Matching formal operand class MCK_FR16 against actual operand at index 3 (Reg:xmm5): match success using generic matcher
 // CHECK:   Matching formal operand class InvalidMatchClass against actual operand at index 4: actual operand index out of range
 // CHECK:   Opcode result: complete match, selecting this opcode
-// CHECK: AsmMatcher: found 2 encodings with mnemonic 'crc32l'
+// CHECK: AsmMatcher: found 4 encodings with mnemonic 'crc32l'
 // CHECK: Trying to match opcode CRC32r32r32
 // CHECK:   Matching formal operand class MCK_GR32 against actual operand at index 1 (Memory: ModeSize=64,BaseReg=rbx,IndexReg=rcx,Scale=8,Disp=2125315823,SegReg=gs): Opcode result: multiple operand mismatches, ignoring this opcode
 // CHECK: Trying to match opcode CRC32r32m32

diff --git a/llvm/test/TableGen/x86-fold-tables.inc b/llvm/test/TableGen/x86-fold-tables.inc
@@ -1602,10 +1602,15 @@ static const X86FoldTableEntry Table2[] = {
   {X86::CMPSSrr, X86::CMPSSrm, 0},
   {X86::CMPSSrr_Int, X86::CMPSSrm_Int, TB_NO_REVERSE},
   {X86::CRC32r32r16, X86::CRC32r32m16, 0},
+  {X86::CRC32r32r16_EVEX, X86::CRC32r32m16_EVEX, 0},
   {X86::CRC32r32r32, X86::CRC32r32m32, 0},
+  {X86::CRC32r32r32_EVEX, X86::CRC32r32m32_EVEX, 0},
   {X86::CRC32r32r8, X86::CRC32r32m8, 0},
+  {X86::CRC32r32r8_EVEX, X86::CRC32r32m8_EVEX, 0},
   {X86::CRC32r64r64, X86::CRC32r64m64, 0},
+  {X86::CRC32r64r64_EVEX, X86::CRC32r64m64_EVEX, 0},
   {X86::CRC32r64r8, X86::CRC32r64m8, 0},
+  {X86::CRC32r64r8_EVEX, X86::CRC32r64m8_EVEX, 0},
   {X86::CVTSD2SSrr_Int, X86::CVTSD2SSrm_Int, TB_NO_REVERSE},
   {X86::CVTSI2SDrr_Int, X86::CVTSI2SDrm_Int, 0},
   {X86::CVTSI2SSrr_Int, X86::CVTSI2SSrm_Int, 0},