From b3625eb44501c241ad711d0f96f954fda8582c8a Mon Sep 17 00:00:00 2001
From: Cameron McInally <cameron.mcinally@nyu.edu>
Date: Fri, 14 Nov 2014 15:43:00 +0000
Subject: [PATCH] [AVX512] Add 512b masked integer shift by immediate patterns.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@222002 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/X86/X86InstrAVX512.td      |  46 ++++------
 lib/Target/X86/X86InstrFormats.td     |   4 +
 test/CodeGen/X86/avx512-intrinsics.ll | 117 ++++++++++++++++++++++----
 3 files changed, 123 insertions(+), 44 deletions(-)

diff --git a/lib/Target/X86/X86InstrAVX512.td b/lib/Target/X86/X86InstrAVX512.td
index 596d2c3bb4b..a9ab0261a59 100644
--- a/lib/Target/X86/X86InstrAVX512.td
+++ b/lib/Target/X86/X86InstrAVX512.td
@@ -3130,29 +3130,17 @@ def : Pat <(i8 (int_x86_avx512_mask_ptestm_q_512 (v8i64 VR512:$src1),
 // AVX-512  Shift instructions
 //===----------------------------------------------------------------------===//
 multiclass avx512_shift_rmi<bits<8> opc, Format ImmFormR, Format ImmFormM,
-                         string OpcodeStr, SDNode OpNode, RegisterClass RC,
-                         ValueType vt, X86MemOperand x86memop, PatFrag mem_frag,
-                         RegisterClass KRC> {
-  def ri : AVX512BIi8<opc, ImmFormR, (outs RC:$dst),
-       (ins RC:$src1, i8imm:$src2),
-           !strconcat(OpcodeStr, " \t{$src2, $src1, $dst|$dst, $src1, $src2}"),
-       [(set RC:$dst, (vt (OpNode RC:$src1, (i8 imm:$src2))))],
-        SSE_INTSHIFT_ITINS_P.rr>, EVEX_4V;
-  def rik : AVX512BIi8<opc, ImmFormR, (outs RC:$dst),
-       (ins KRC:$mask, RC:$src1, i8imm:$src2),
-           !strconcat(OpcodeStr,
-                " \t{$src2, $src1, $dst {${mask}}|$dst {${mask}}, $src1, $src2}"),
-       [], SSE_INTSHIFT_ITINS_P.rr>, EVEX_4V, EVEX_K;
-  def mi: AVX512BIi8<opc, ImmFormM, (outs RC:$dst),
-       (ins x86memop:$src1, i8imm:$src2),
-           !strconcat(OpcodeStr, " \t{$src2, $src1, $dst|$dst, $src1, $src2}"),
-       [(set RC:$dst, (OpNode (mem_frag addr:$src1),
-                     (i8 imm:$src2)))], SSE_INTSHIFT_ITINS_P.rm>, EVEX_4V;
-  def mik: AVX512BIi8<opc, ImmFormM, (outs RC:$dst),
-       (ins KRC:$mask, x86memop:$src1, i8imm:$src2),
-           !strconcat(OpcodeStr,
-                " \t{$src2, $src1, $dst {${mask}}|$dst {${mask}}, $src1, $src2}"),
-       [], SSE_INTSHIFT_ITINS_P.rm>, EVEX_4V, EVEX_K;
+                         string OpcodeStr, SDNode OpNode, X86VectorVTInfo _> { 
+  defm ri : AVX512_maskable<opc, ImmFormR, _, (outs _.RC:$dst),
+                   (ins _.RC:$src1, i8imm:$src2), OpcodeStr,
+                      "$src2, $src1", "$src1, $src2",
+                   (_.VT (OpNode _.RC:$src1, (i8 imm:$src2))),
+                   " ",  SSE_INTSHIFT_ITINS_P.rr>, AVX512BIi8Base, EVEX_4V;
+  defm mi : AVX512_maskable<opc, ImmFormM, _, (outs _.RC:$dst),
+                   (ins _.MemOp:$src1, i8imm:$src2), OpcodeStr,
+                       "$src2, $src1", "$src1, $src2",
+                   (_.VT (OpNode (_.MemOpFrag addr:$src1), (i8 imm:$src2))),
+                   " ",  SSE_INTSHIFT_ITINS_P.rm>, AVX512BIi8Base, EVEX_4V;
 }
 
 multiclass avx512_shift_rrm<bits<8> opc, string OpcodeStr, SDNode OpNode,
@@ -3183,42 +3171,42 @@ multiclass avx512_shift_rrm<bits<8> opc, string OpcodeStr, SDNode OpNode,
 }
 
 defm VPSRLDZ : avx512_shift_rmi<0x72, MRM2r, MRM2m, "vpsrld", X86vsrli,
-                           VR512, v16i32, i512mem, memopv16i32, VK16WM>,
+                           v16i32_info>,
                            EVEX_V512, EVEX_CD8<32, CD8VF>;
 defm VPSRLDZ : avx512_shift_rrm<0xD2, "vpsrld", X86vsrl,
                            VR512, v16i32, v4i32, bc_v4i32, VK16WM>, EVEX_V512,
                            EVEX_CD8<32, CD8VQ>;
                            
 defm VPSRLQZ : avx512_shift_rmi<0x73, MRM2r, MRM2m, "vpsrlq", X86vsrli,
-                           VR512, v8i64, i512mem, memopv8i64, VK8WM>, EVEX_V512,
+                           v8i64_info>, EVEX_V512,
                            EVEX_CD8<64, CD8VF>, VEX_W;
 defm VPSRLQZ : avx512_shift_rrm<0xD3, "vpsrlq", X86vsrl,
                            VR512, v8i64, v2i64, bc_v2i64, VK8WM>, EVEX_V512,
                            EVEX_CD8<64, CD8VQ>, VEX_W;
 
 defm VPSLLDZ : avx512_shift_rmi<0x72, MRM6r, MRM6m, "vpslld", X86vshli,
-                           VR512, v16i32, i512mem, memopv16i32, VK16WM>, EVEX_V512,
+                           v16i32_info>, EVEX_V512,
                            EVEX_CD8<32, CD8VF>;
 defm VPSLLDZ : avx512_shift_rrm<0xF2, "vpslld", X86vshl,
                            VR512, v16i32, v4i32, bc_v4i32, VK16WM>, EVEX_V512,
                            EVEX_CD8<32, CD8VQ>;
                            
 defm VPSLLQZ : avx512_shift_rmi<0x73, MRM6r, MRM6m, "vpsllq", X86vshli,
-                           VR512, v8i64, i512mem, memopv8i64, VK8WM>, EVEX_V512,
+                           v8i64_info>, EVEX_V512,
                            EVEX_CD8<64, CD8VF>, VEX_W;
 defm VPSLLQZ : avx512_shift_rrm<0xF3, "vpsllq", X86vshl,
                            VR512, v8i64, v2i64, bc_v2i64, VK8WM>, EVEX_V512,
                            EVEX_CD8<64, CD8VQ>, VEX_W;
 
 defm VPSRADZ : avx512_shift_rmi<0x72, MRM4r, MRM4m, "vpsrad", X86vsrai,
-                           VR512, v16i32, i512mem, memopv16i32, VK16WM>,
+                           v16i32_info>,
                            EVEX_V512, EVEX_CD8<32, CD8VF>;
 defm VPSRADZ : avx512_shift_rrm<0xE2, "vpsrad", X86vsra,
                            VR512, v16i32, v4i32, bc_v4i32, VK16WM>, EVEX_V512,
                            EVEX_CD8<32, CD8VQ>;
                            
 defm VPSRAQZ : avx512_shift_rmi<0x72, MRM4r, MRM4m, "vpsraq", X86vsrai,
-                           VR512, v8i64, i512mem, memopv8i64, VK8WM>, EVEX_V512,
+                           v8i64_info>, EVEX_V512,
                            EVEX_CD8<64, CD8VF>, VEX_W;
 defm VPSRAQZ : avx512_shift_rrm<0xE2, "vpsraq", X86vsra,
                            VR512, v8i64, v2i64, bc_v2i64, VK8WM>, EVEX_V512,
diff --git a/lib/Target/X86/X86InstrFormats.td b/lib/Target/X86/X86InstrFormats.td
index cbd135aa007..fe4ead1905c 100644
--- a/lib/Target/X86/X86InstrFormats.td
+++ b/lib/Target/X86/X86InstrFormats.td
@@ -722,6 +722,10 @@ class AVX512BIi8<bits<8> o, Format F, dag outs, dag ins, string asm,
               list<dag> pattern, InstrItinClass itin = NoItinerary>
       : Ii8<o, F, outs, ins, asm, pattern, itin, SSEPackedInt>, PD,
         Requires<[HasAVX512]>;
+class AVX512BIi8Base : PD {
+  Domain ExeDomain = SSEPackedInt;
+  ImmType ImmT = Imm8;
+}
 class AVX512AIi8<bits<8> o, Format F, dag outs, dag ins, string asm,
               list<dag> pattern, InstrItinClass itin = NoItinerary>
       : Ii8<o, F, outs, ins, asm, pattern, itin, SSEPackedInt>, TAPD,
diff --git a/test/CodeGen/X86/avx512-intrinsics.ll b/test/CodeGen/X86/avx512-intrinsics.ll
index 11d12c09fea..691d1fbc1d4 100644
--- a/test/CodeGen/X86/avx512-intrinsics.ll
+++ b/test/CodeGen/X86/avx512-intrinsics.ll
@@ -951,53 +951,140 @@ define <4 x double> @test_vextractf64x4(<8 x double> %a) {
 
 declare <4 x double> @llvm.x86.avx512.mask.vextractf64x4.512(<8 x double>, i8, <4 x double>, i8)
 
-define <16 x i32> @test_x86_avx512_mask_pslli_d(<16 x i32> %a0) {
-  ; CHECK-LABEL: test_x86_avx512_mask_pslli_d 
+define <16 x i32> @test_x86_avx512_pslli_d(<16 x i32> %a0) {
+  ; CHECK-LABEL: test_x86_avx512_pslli_d
   ; CHECK: vpslld
   %res = call <16 x i32> @llvm.x86.avx512.mask.pslli.d(<16 x i32> %a0, i32 7, <16 x i32> zeroinitializer, i16 -1)
   ret <16 x i32> %res
 }
+
+define <16 x i32> @test_x86_avx512_mask_pslli_d(<16 x i32> %a0, <16 x i32> %a1, i16 %mask) {
+  ; CHECK-LABEL: test_x86_avx512_mask_pslli_d 
+  ; CHECK: vpslld $7, %zmm0, %zmm1 {%k1}  
+  %res = call <16 x i32> @llvm.x86.avx512.mask.pslli.d(<16 x i32> %a0, i32 7, <16 x i32> %a1, i16 %mask)
+  ret <16 x i32> %res
+}
+
+define <16 x i32> @test_x86_avx512_maskz_pslli_d(<16 x i32> %a0, i16 %mask) {
+  ; CHECK-LABEL: test_x86_avx512_maskz_pslli_d
+  ; CHECK: vpslld $7, %zmm0, %zmm0 {%k1} {z}
+  %res = call <16 x i32> @llvm.x86.avx512.mask.pslli.d(<16 x i32> %a0, i32 7, <16 x i32> zeroinitializer, i16 %mask)
+  ret <16 x i32> %res
+}
+
 declare <16 x i32> @llvm.x86.avx512.mask.pslli.d(<16 x i32>, i32, <16 x i32>, i16) nounwind readnone
 
-
-define <8 x i64> @test_x86_avx512_mask_pslli_q(<8 x i64> %a0) {
-  ; CHECK-LABEL: test_x86_avx512_mask_pslli_q
+define <8 x i64> @test_x86_avx512_pslli_q(<8 x i64> %a0) {
+  ; CHECK-LABEL: test_x86_avx512_pslli_q
   ; CHECK: vpsllq
   %res = call <8 x i64> @llvm.x86.avx512.mask.pslli.q(<8 x i64> %a0, i32 7, <8 x i64> zeroinitializer, i8 -1)
   ret <8 x i64> %res
 }
+
+define <8 x i64> @test_x86_avx512_mask_pslli_q(<8 x i64> %a0, <8 x i64> %a1, i8 %mask) {
+  ; CHECK-LABEL: test_x86_avx512_mask_pslli_q
+  ; CHECK: vpsllq $7, %zmm0, %zmm1 {%k1}   
+  %res = call <8 x i64> @llvm.x86.avx512.mask.pslli.q(<8 x i64> %a0, i32 7, <8 x i64> %a1, i8 %mask)
+  ret <8 x i64> %res
+}
+
+define <8 x i64> @test_x86_avx512_maskz_pslli_q(<8 x i64> %a0, i8 %mask) {
+  ; CHECK-LABEL: test_x86_avx512_maskz_pslli_q
+  ; CHECK: vpsllq $7, %zmm0, %zmm0 {%k1} {z} 
+  %res = call <8 x i64> @llvm.x86.avx512.mask.pslli.q(<8 x i64> %a0, i32 7, <8 x i64> zeroinitializer, i8 %mask)
+  ret <8 x i64> %res
+}
+
 declare <8 x i64> @llvm.x86.avx512.mask.pslli.q(<8 x i64>, i32, <8 x i64>, i8) nounwind readnone
 
-define <16 x i32> @test_x86_avx512_mask_psrli_d(<16 x i32> %a0) {
-  ; CHECK-LABEL: test_x86_avx512_mask_psrli_d
+define <16 x i32> @test_x86_avx512_psrli_d(<16 x i32> %a0) {
+  ; CHECK-LABEL: test_x86_avx512_psrli_d
   ; CHECK: vpsrld
   %res = call <16 x i32> @llvm.x86.avx512.mask.psrli.d(<16 x i32> %a0, i32 7, <16 x i32> zeroinitializer, i16 -1)
   ret <16 x i32> %res
 }
+
+define <16 x i32> @test_x86_avx512_mask_psrli_d(<16 x i32> %a0, <16 x i32> %a1, i16 %mask) {
+  ; CHECK-LABEL: test_x86_avx512_mask_psrli_d
+  ; CHECK: vpsrld $7, %zmm0, %zmm1 {%k1}  
+  %res = call <16 x i32> @llvm.x86.avx512.mask.psrli.d(<16 x i32> %a0, i32 7, <16 x i32> %a1, i16 %mask)
+  ret <16 x i32> %res
+}
+
+define <16 x i32> @test_x86_avx512_maskz_psrli_d(<16 x i32> %a0, i16 %mask) {
+  ; CHECK-LABEL: test_x86_avx512_maskz_psrli_d
+  ; CHECK: vpsrld $7, %zmm0, %zmm0 {%k1} {z}
+  %res = call <16 x i32> @llvm.x86.avx512.mask.psrli.d(<16 x i32> %a0, i32 7, <16 x i32> zeroinitializer, i16 %mask)
+  ret <16 x i32> %res
+}
+
 declare <16 x i32> @llvm.x86.avx512.mask.psrli.d(<16 x i32>, i32, <16 x i32>, i16) nounwind readnone
 
-
-define <8 x i64> @test_x86_avx512_mask_psrli_q(<8 x i64> %a0) {
-  ; CHECK-LABEL: test_x86_avx512_mask_psrli_q
+define <8 x i64> @test_x86_avx512_psrli_q(<8 x i64> %a0) {
+  ; CHECK-LABEL: test_x86_avx512_psrli_q
   ; CHECK: vpsrlq
   %res = call <8 x i64> @llvm.x86.avx512.mask.psrli.q(<8 x i64> %a0, i32 7, <8 x i64> zeroinitializer, i8 -1)
   ret <8 x i64> %res
 }
+
+define <8 x i64> @test_x86_avx512_mask_psrli_q(<8 x i64> %a0, <8 x i64> %a1, i8 %mask) {
+  ; CHECK-LABEL: test_x86_avx512_mask_psrli_q
+  ; CHECK: vpsrlq $7, %zmm0, %zmm1 {%k1}  
+  %res = call <8 x i64> @llvm.x86.avx512.mask.psrli.q(<8 x i64> %a0, i32 7, <8 x i64> %a1, i8 %mask)
+  ret <8 x i64> %res
+}
+
+define <8 x i64> @test_x86_avx512_maskz_psrli_q(<8 x i64> %a0, i8 %mask) {
+  ; CHECK-LABEL: test_x86_avx512_maskz_psrli_q
+  ; CHECK: vpsrlq $7, %zmm0, %zmm0 {%k1} {z}
+  %res = call <8 x i64> @llvm.x86.avx512.mask.psrli.q(<8 x i64> %a0, i32 7, <8 x i64> zeroinitializer, i8 %mask)
+  ret <8 x i64> %res
+}
+
 declare <8 x i64> @llvm.x86.avx512.mask.psrli.q(<8 x i64>, i32, <8 x i64>, i8) nounwind readnone
 
-define <16 x i32> @test_x86_avx512_mask_psrai_d(<16 x i32> %a0) {
-  ; CHECK-LABEL: test_x86_avx512_mask_psrai_d
+define <16 x i32> @test_x86_avx512_psrai_d(<16 x i32> %a0) {
+  ; CHECK-LABEL: test_x86_avx512_psrai_d
   ; CHECK: vpsrad
   %res = call <16 x i32> @llvm.x86.avx512.mask.psrai.d(<16 x i32> %a0, i32 7, <16 x i32> zeroinitializer, i16 -1)
   ret <16 x i32> %res
 }
+
+define <16 x i32> @test_x86_avx512_mask_psrai_d(<16 x i32> %a0, <16 x i32> %a1, i16 %mask) {
+  ; CHECK-LABEL: test_x86_avx512_mask_psrai_d
+  ; CHECK: vpsrad $7, %zmm0, %zmm1 {%k1}  
+  %res = call <16 x i32> @llvm.x86.avx512.mask.psrai.d(<16 x i32> %a0, i32 7, <16 x i32> %a1, i16 %mask)
+  ret <16 x i32> %res
+}
+
+define <16 x i32> @test_x86_avx512_maskz_psrai_d(<16 x i32> %a0, i16 %mask) {
+  ; CHECK-LABEL: test_x86_avx512_maskz_psrai_d
+  ; CHECK: vpsrad $7, %zmm0, %zmm0 {%k1} {z}
+  %res = call <16 x i32> @llvm.x86.avx512.mask.psrai.d(<16 x i32> %a0, i32 7, <16 x i32> zeroinitializer, i16 %mask)
+  ret <16 x i32> %res
+}
+
 declare <16 x i32> @llvm.x86.avx512.mask.psrai.d(<16 x i32>, i32, <16 x i32>, i16) nounwind readnone
 
-
-define <8 x i64> @test_x86_avx512_mask_psrai_q(<8 x i64> %a0) {
-  ; CHECK-LABEL: test_x86_avx512_mask_psrai_q
+define <8 x i64> @test_x86_avx512_psrai_q(<8 x i64> %a0) {
+  ; CHECK-LABEL: test_x86_avx512_psrai_q
   ; CHECK: vpsraq
   %res = call <8 x i64> @llvm.x86.avx512.mask.psrai.q(<8 x i64> %a0, i32 7, <8 x i64> zeroinitializer, i8 -1)
   ret <8 x i64> %res
 }
+
+define <8 x i64> @test_x86_avx512_mask_psrai_q(<8 x i64> %a0, <8 x i64> %a1, i8 %mask) {
+  ; CHECK-LABEL: test_x86_avx512_mask_psrai_q
+  ; CHECK: vpsraq $7, %zmm0, %zmm1 {%k1}   
+  %res = call <8 x i64> @llvm.x86.avx512.mask.psrai.q(<8 x i64> %a0, i32 7, <8 x i64> %a1, i8 %mask)
+  ret <8 x i64> %res
+}
+
+define <8 x i64> @test_x86_avx512_maskz_psrai_q(<8 x i64> %a0, i8 %mask) {
+  ; CHECK-LABEL: test_x86_avx512_maskz_psrai_q
+  ; CHECK: vpsraq $7, %zmm0, %zmm0 {%k1} {z}
+  %res = call <8 x i64> @llvm.x86.avx512.mask.psrai.q(<8 x i64> %a0, i32 7, <8 x i64> zeroinitializer, i8 %mask)
+  ret <8 x i64> %res
+}
+
 declare <8 x i64> @llvm.x86.avx512.mask.psrai.q(<8 x i64>, i32, <8 x i64>, i8) nounwind readnone