From a3ea5549ccb262948357826015158cd4d245fc9d Mon Sep 17 00:00:00 2001
From: Simon Pilgrim <llvm-dev@redking.me.uk>
Date: Thu, 7 Dec 2017 15:24:14 +0000
Subject: [PATCH] [X86] Tag LZCNT/TZCNT instructions scheduler classes

Tagged as IMUL instructions for a reasonable approximation (ALU tends to be a lot faster) - POPCNT is currently tagged as FAdd which I think should be replaced with IMUL as well

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@320051 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/X86/X86InstrInfo.td     | 42 ++++++++++++++++++++++----------------
 lib/Target/X86/X86Schedule.td      |  4 ++++
 test/CodeGen/X86/bmi-schedule.ll   | 24 +++++++++++-----------
 test/CodeGen/X86/lzcnt-schedule.ll | 24 +++++++++++-----------
 4 files changed, 52 insertions(+), 42 deletions(-)

diff --git a/lib/Target/X86/X86InstrInfo.td b/lib/Target/X86/X86InstrInfo.td
index dafec20bc09..0931a9c98d1 100644
--- a/lib/Target/X86/X86InstrInfo.td
+++ b/lib/Target/X86/X86InstrInfo.td
@@ -2235,30 +2235,33 @@ let Predicates = [HasRDSEED], Defs = [EFLAGS], SchedRW = [WriteSystem] in {
 let Predicates = [HasLZCNT], Defs = [EFLAGS] in {
   def LZCNT16rr : I<0xBD, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
                     "lzcnt{w}\t{$src, $dst|$dst, $src}",
-                    [(set GR16:$dst, (ctlz GR16:$src)), (implicit EFLAGS)]>, XS,
-                    OpSize16;
+                    [(set GR16:$dst, (ctlz GR16:$src)), (implicit EFLAGS)],
+                    IIC_LZCNT_RR>, XS, OpSize16, Sched<[WriteIMul]>;
   def LZCNT16rm : I<0xBD, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
                     "lzcnt{w}\t{$src, $dst|$dst, $src}",
                     [(set GR16:$dst, (ctlz (loadi16 addr:$src))),
-                     (implicit EFLAGS)]>, XS, OpSize16;
+                     (implicit EFLAGS)], IIC_LZCNT_RM>, XS, OpSize16,
+                    Sched<[WriteIMulLd]>;
 
   def LZCNT32rr : I<0xBD, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
                     "lzcnt{l}\t{$src, $dst|$dst, $src}",
-                    [(set GR32:$dst, (ctlz GR32:$src)), (implicit EFLAGS)]>, XS,
-                    OpSize32;
+                    [(set GR32:$dst, (ctlz GR32:$src)), (implicit EFLAGS)],
+                    IIC_LZCNT_RR>, XS, OpSize32, Sched<[WriteIMul]>;
   def LZCNT32rm : I<0xBD, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
                     "lzcnt{l}\t{$src, $dst|$dst, $src}",
                     [(set GR32:$dst, (ctlz (loadi32 addr:$src))),
-                     (implicit EFLAGS)]>, XS, OpSize32;
+                     (implicit EFLAGS)], IIC_LZCNT_RM>, XS, OpSize32,
+                    Sched<[WriteIMulLd]>;
 
   def LZCNT64rr : RI<0xBD, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
                      "lzcnt{q}\t{$src, $dst|$dst, $src}",
-                     [(set GR64:$dst, (ctlz GR64:$src)), (implicit EFLAGS)]>,
-                     XS;
+                     [(set GR64:$dst, (ctlz GR64:$src)), (implicit EFLAGS)],
+                     IIC_LZCNT_RR>, XS, Sched<[WriteIMul]>;
   def LZCNT64rm : RI<0xBD, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
                      "lzcnt{q}\t{$src, $dst|$dst, $src}",
                      [(set GR64:$dst, (ctlz (loadi64 addr:$src))),
-                      (implicit EFLAGS)]>, XS;
+                      (implicit EFLAGS)], IIC_LZCNT_RM>, XS,
+                     Sched<[WriteIMulLd]>;
 }
 
 //===----------------------------------------------------------------------===//
@@ -2267,30 +2270,33 @@ let Predicates = [HasLZCNT], Defs = [EFLAGS] in {
 let Predicates = [HasBMI], Defs = [EFLAGS] in {
   def TZCNT16rr : I<0xBC, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
                     "tzcnt{w}\t{$src, $dst|$dst, $src}",
-                    [(set GR16:$dst, (cttz GR16:$src)), (implicit EFLAGS)]>, XS,
-                    OpSize16;
+                    [(set GR16:$dst, (cttz GR16:$src)), (implicit EFLAGS)],
+                    IIC_TZCNT_RR>, XS, OpSize16, Sched<[WriteIMul]>;
   def TZCNT16rm : I<0xBC, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
                     "tzcnt{w}\t{$src, $dst|$dst, $src}",
                     [(set GR16:$dst, (cttz (loadi16 addr:$src))),
-                     (implicit EFLAGS)]>, XS, OpSize16;
+                     (implicit EFLAGS)], IIC_TZCNT_RM>, XS, OpSize16,
+                    Sched<[WriteIMulLd]>;
 
   def TZCNT32rr : I<0xBC, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
                     "tzcnt{l}\t{$src, $dst|$dst, $src}",
-                    [(set GR32:$dst, (cttz GR32:$src)), (implicit EFLAGS)]>, XS,
-                    OpSize32;
+                    [(set GR32:$dst, (cttz GR32:$src)), (implicit EFLAGS)],
+                    IIC_TZCNT_RR>, XS, OpSize32, Sched<[WriteIMul]>;
   def TZCNT32rm : I<0xBC, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
                     "tzcnt{l}\t{$src, $dst|$dst, $src}",
                     [(set GR32:$dst, (cttz (loadi32 addr:$src))),
-                     (implicit EFLAGS)]>, XS, OpSize32;
+                     (implicit EFLAGS)], IIC_TZCNT_RM>, XS, OpSize32,
+                    Sched<[WriteIMulLd]>;
 
   def TZCNT64rr : RI<0xBC, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
                      "tzcnt{q}\t{$src, $dst|$dst, $src}",
-                     [(set GR64:$dst, (cttz GR64:$src)), (implicit EFLAGS)]>,
-                     XS;
+                     [(set GR64:$dst, (cttz GR64:$src)), (implicit EFLAGS)],
+                     IIC_TZCNT_RR>, XS, Sched<[WriteIMul]>;
   def TZCNT64rm : RI<0xBC, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
                      "tzcnt{q}\t{$src, $dst|$dst, $src}",
                      [(set GR64:$dst, (cttz (loadi64 addr:$src))),
-                      (implicit EFLAGS)]>, XS;
+                      (implicit EFLAGS)], IIC_TZCNT_RM>, XS,
+                     Sched<[WriteIMulLd]>;
 }
 
 multiclass bmi_bls<string mnemonic, Format RegMRM, Format MemMRM,
diff --git a/lib/Target/X86/X86Schedule.td b/lib/Target/X86/X86Schedule.td
index 18410d6227e..e35c1decc7c 100644
--- a/lib/Target/X86/X86Schedule.td
+++ b/lib/Target/X86/X86Schedule.td
@@ -564,6 +564,10 @@ def IIC_PUSH_A : InstrItinClass;
 def IIC_BSWAP : InstrItinClass;
 def IIC_BIT_SCAN_MEM : InstrItinClass;
 def IIC_BIT_SCAN_REG : InstrItinClass;
+def IIC_LZCNT_RR : InstrItinClass;
+def IIC_LZCNT_RM : InstrItinClass;
+def IIC_TZCNT_RR : InstrItinClass;
+def IIC_TZCNT_RM : InstrItinClass;
 def IIC_MOVS : InstrItinClass;
 def IIC_STOS : InstrItinClass;
 def IIC_SCAS : InstrItinClass;
diff --git a/test/CodeGen/X86/bmi-schedule.ll b/test/CodeGen/X86/bmi-schedule.ll
index 59817ad504a..fcbe66448ad 100644
--- a/test/CodeGen/X86/bmi-schedule.ll
+++ b/test/CodeGen/X86/bmi-schedule.ll
@@ -578,8 +578,8 @@ define i64 @test_blsr_i64(i64 %a0, i64 *%a1) {
 define i16 @test_cttz_i16(i16 zeroext %a0, i16 *%a1) {
 ; GENERIC-LABEL: test_cttz_i16:
 ; GENERIC:       # %bb.0:
-; GENERIC-NEXT:    tzcntw (%rsi), %cx
-; GENERIC-NEXT:    tzcntw %di, %ax
+; GENERIC-NEXT:    tzcntw (%rsi), %cx # sched: [7:1.00]
+; GENERIC-NEXT:    tzcntw %di, %ax # sched: [3:1.00]
 ; GENERIC-NEXT:    orl %ecx, %eax # sched: [1:0.33]
 ; GENERIC-NEXT:    # kill: def %ax killed %ax killed %eax
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
@@ -610,8 +610,8 @@ define i16 @test_cttz_i16(i16 zeroext %a0, i16 *%a1) {
 ;
 ; BTVER2-LABEL: test_cttz_i16:
 ; BTVER2:       # %bb.0:
-; BTVER2-NEXT:    tzcntw (%rsi), %cx
-; BTVER2-NEXT:    tzcntw %di, %ax
+; BTVER2-NEXT:    tzcntw (%rsi), %cx # sched: [6:1.00]
+; BTVER2-NEXT:    tzcntw %di, %ax # sched: [3:1.00]
 ; BTVER2-NEXT:    orl %ecx, %eax # sched: [1:0.50]
 ; BTVER2-NEXT:    # kill: def %ax killed %ax killed %eax
 ; BTVER2-NEXT:    retq # sched: [4:1.00]
@@ -634,8 +634,8 @@ declare i16 @llvm.cttz.i16(i16, i1)
 define i32 @test_cttz_i32(i32 %a0, i32 *%a1) {
 ; GENERIC-LABEL: test_cttz_i32:
 ; GENERIC:       # %bb.0:
-; GENERIC-NEXT:    tzcntl (%rsi), %ecx
-; GENERIC-NEXT:    tzcntl %edi, %eax
+; GENERIC-NEXT:    tzcntl (%rsi), %ecx # sched: [7:1.00]
+; GENERIC-NEXT:    tzcntl %edi, %eax # sched: [3:1.00]
 ; GENERIC-NEXT:    orl %ecx, %eax # sched: [1:0.33]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
@@ -662,8 +662,8 @@ define i32 @test_cttz_i32(i32 %a0, i32 *%a1) {
 ;
 ; BTVER2-LABEL: test_cttz_i32:
 ; BTVER2:       # %bb.0:
-; BTVER2-NEXT:    tzcntl (%rsi), %ecx
-; BTVER2-NEXT:    tzcntl %edi, %eax
+; BTVER2-NEXT:    tzcntl (%rsi), %ecx # sched: [6:1.00]
+; BTVER2-NEXT:    tzcntl %edi, %eax # sched: [3:1.00]
 ; BTVER2-NEXT:    orl %ecx, %eax # sched: [1:0.50]
 ; BTVER2-NEXT:    retq # sched: [4:1.00]
 ;
@@ -684,8 +684,8 @@ declare i32 @llvm.cttz.i32(i32, i1)
 define i64 @test_cttz_i64(i64 %a0, i64 *%a1) {
 ; GENERIC-LABEL: test_cttz_i64:
 ; GENERIC:       # %bb.0:
-; GENERIC-NEXT:    tzcntq (%rsi), %rcx
-; GENERIC-NEXT:    tzcntq %rdi, %rax
+; GENERIC-NEXT:    tzcntq (%rsi), %rcx # sched: [7:1.00]
+; GENERIC-NEXT:    tzcntq %rdi, %rax # sched: [3:1.00]
 ; GENERIC-NEXT:    orq %rcx, %rax # sched: [1:0.33]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
@@ -712,8 +712,8 @@ define i64 @test_cttz_i64(i64 %a0, i64 *%a1) {
 ;
 ; BTVER2-LABEL: test_cttz_i64:
 ; BTVER2:       # %bb.0:
-; BTVER2-NEXT:    tzcntq (%rsi), %rcx
-; BTVER2-NEXT:    tzcntq %rdi, %rax
+; BTVER2-NEXT:    tzcntq (%rsi), %rcx # sched: [6:1.00]
+; BTVER2-NEXT:    tzcntq %rdi, %rax # sched: [3:1.00]
 ; BTVER2-NEXT:    orq %rcx, %rax # sched: [1:0.50]
 ; BTVER2-NEXT:    retq # sched: [4:1.00]
 ;
diff --git a/test/CodeGen/X86/lzcnt-schedule.ll b/test/CodeGen/X86/lzcnt-schedule.ll
index b56bd555e78..2aeb79eefd7 100644
--- a/test/CodeGen/X86/lzcnt-schedule.ll
+++ b/test/CodeGen/X86/lzcnt-schedule.ll
@@ -10,8 +10,8 @@
 define i16 @test_ctlz_i16(i16 zeroext %a0, i16 *%a1) {
 ; GENERIC-LABEL: test_ctlz_i16:
 ; GENERIC:       # %bb.0:
-; GENERIC-NEXT:    lzcntw (%rsi), %cx
-; GENERIC-NEXT:    lzcntw %di, %ax
+; GENERIC-NEXT:    lzcntw (%rsi), %cx # sched: [7:1.00]
+; GENERIC-NEXT:    lzcntw %di, %ax # sched: [3:1.00]
 ; GENERIC-NEXT:    orl %ecx, %eax # sched: [1:0.33]
 ; GENERIC-NEXT:    # kill: def %ax killed %ax killed %eax
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
@@ -42,8 +42,8 @@ define i16 @test_ctlz_i16(i16 zeroext %a0, i16 *%a1) {
 ;
 ; BTVER2-LABEL: test_ctlz_i16:
 ; BTVER2:       # %bb.0:
-; BTVER2-NEXT:    lzcntw (%rsi), %cx
-; BTVER2-NEXT:    lzcntw %di, %ax
+; BTVER2-NEXT:    lzcntw (%rsi), %cx # sched: [6:1.00]
+; BTVER2-NEXT:    lzcntw %di, %ax # sched: [3:1.00]
 ; BTVER2-NEXT:    orl %ecx, %eax # sched: [1:0.50]
 ; BTVER2-NEXT:    # kill: def %ax killed %ax killed %eax
 ; BTVER2-NEXT:    retq # sched: [4:1.00]
@@ -66,8 +66,8 @@ declare i16 @llvm.ctlz.i16(i16, i1)
 define i32 @test_ctlz_i32(i32 %a0, i32 *%a1) {
 ; GENERIC-LABEL: test_ctlz_i32:
 ; GENERIC:       # %bb.0:
-; GENERIC-NEXT:    lzcntl (%rsi), %ecx
-; GENERIC-NEXT:    lzcntl %edi, %eax
+; GENERIC-NEXT:    lzcntl (%rsi), %ecx # sched: [7:1.00]
+; GENERIC-NEXT:    lzcntl %edi, %eax # sched: [3:1.00]
 ; GENERIC-NEXT:    orl %ecx, %eax # sched: [1:0.33]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
@@ -94,8 +94,8 @@ define i32 @test_ctlz_i32(i32 %a0, i32 *%a1) {
 ;
 ; BTVER2-LABEL: test_ctlz_i32:
 ; BTVER2:       # %bb.0:
-; BTVER2-NEXT:    lzcntl (%rsi), %ecx
-; BTVER2-NEXT:    lzcntl %edi, %eax
+; BTVER2-NEXT:    lzcntl (%rsi), %ecx # sched: [6:1.00]
+; BTVER2-NEXT:    lzcntl %edi, %eax # sched: [3:1.00]
 ; BTVER2-NEXT:    orl %ecx, %eax # sched: [1:0.50]
 ; BTVER2-NEXT:    retq # sched: [4:1.00]
 ;
@@ -116,8 +116,8 @@ declare i32 @llvm.ctlz.i32(i32, i1)
 define i64 @test_ctlz_i64(i64 %a0, i64 *%a1) {
 ; GENERIC-LABEL: test_ctlz_i64:
 ; GENERIC:       # %bb.0:
-; GENERIC-NEXT:    lzcntq (%rsi), %rcx
-; GENERIC-NEXT:    lzcntq %rdi, %rax
+; GENERIC-NEXT:    lzcntq (%rsi), %rcx # sched: [7:1.00]
+; GENERIC-NEXT:    lzcntq %rdi, %rax # sched: [3:1.00]
 ; GENERIC-NEXT:    orq %rcx, %rax # sched: [1:0.33]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
@@ -144,8 +144,8 @@ define i64 @test_ctlz_i64(i64 %a0, i64 *%a1) {
 ;
 ; BTVER2-LABEL: test_ctlz_i64:
 ; BTVER2:       # %bb.0:
-; BTVER2-NEXT:    lzcntq (%rsi), %rcx
-; BTVER2-NEXT:    lzcntq %rdi, %rax
+; BTVER2-NEXT:    lzcntq (%rsi), %rcx # sched: [6:1.00]
+; BTVER2-NEXT:    lzcntq %rdi, %rax # sched: [3:1.00]
 ; BTVER2-NEXT:    orq %rcx, %rax # sched: [1:0.50]
 ; BTVER2-NEXT:    retq # sched: [4:1.00]
 ;
-- 
2.11.0