Skip to content

Commit 4f91e5c

Browse files
mingmingl-llvmmingmingl-llvm
committed
resolve comments
1 parent 23fd549 commit 4f91e5c

File tree

2 files changed

+131
-90
lines changed

2 files changed

+131
-90
lines changed

‎llvm/test/CodeGen/AArch64/constant-pool-partition.ll‎

Lines changed: 82 additions & 51 deletions
Original file line numberDiff line numberDiff line change
@@ -14,56 +14,84 @@
1414
; secondly, and then hot_func. Specifically, tests that
1515
; - If a constant is accessed by hot functions, all constant pools for this
1616
; constant (e.g., from an unprofiled function, or cold function) should have
17-
; `.hot` suffix.
17+
; `.hot` suffix. For instance, double 0.68 is seen by both @cold_func and
18+
; @hot_func, so two CPI emits (under label LCPI0_0 and LCPI2_0) have `.hot`
19+
; suffix.
1820
; - Similarly if a constant is accessed by both cold function and un-profiled
1921
; function, constant pools for this constant should not have `.unlikely` suffix.
2022

21-
; CHECK: .section .rodata.cst8.hot,"aM",@progbits,8
22-
; CHECK: .LCPI0_0:
23-
; CHECK: .xword 0x3fe5c28f5c28f5c3 // double 0.68000000000000005
24-
; CHECK: .section .rodata.cst8.unlikely,"aM",@progbits,8
25-
; CHECK: .LCPI0_1:
26-
; CHECK: .xword 0x3fe5eb851eb851ec // double 0.68500000000000005
27-
; CHECK: .section .rodata.cst8,"aM",@progbits,8
28-
; CHECK: .LCPI0_2:
29-
; CHECK: .byte 0 // 0x0
30-
; CHECK: .byte 4 // 0x4
31-
; CHECK: .byte 8 // 0x8
32-
; CHECK: .byte 12 // 0xc
33-
; CHECK: .byte 255 // 0xff
34-
; CHECK: .byte 255 // 0xff
35-
; CHECK: .byte 255 // 0xff
36-
; CHECK: .byte 255 // 0xff
37-
38-
; CHECK: .section .rodata.cst8,"aM",@progbits,8
39-
; CHECK: .LCPI1_0:
40-
; CHECK: .byte 0 // 0x0
41-
; CHECK: .byte 4 // 0x4
42-
; CHECK: .byte 8 // 0x8
43-
; CHECK: .byte 12 // 0xc
44-
; CHECK: .byte 255 // 0xff
45-
; CHECK: .byte 255 // 0xff
46-
; CHECK: .byte 255 // 0xff
47-
; CHECK: .byte 255 // 0xff
48-
; CHECK: .section .rodata.cst16.hot,"aM",@progbits,16
49-
; CHECK: .LCPI1_1:
50-
; CHECK: .word 442 // 0x1ba
51-
; CHECK: .word 100 // 0x64
52-
; CHECK: .word 0 // 0x0
53-
; CHECK: .word 0 // 0x0
54-
23+
;; Constant pools for function @cold_func.
24+
; CHECK: .section .rodata.cst8.hot,"aM",@progbits,8
25+
; CHECK-NEXT: .p2align
26+
; CHECK-NEXT: .LCPI0_0:
27+
; CHECK-NEXT: .xword 0x3fe5c28f5c28f5c3 // double 0.68000000000000005
28+
; CHECK-NEXT: .section .rodata.cst8.unlikely,"aM",@progbits,8
29+
; CHECK-NEXT: .p2align
30+
; CHECK-NEXT: .LCPI0_1:
31+
; CHECK-NEXT: .xword 0x3fe5eb851eb851ec // double 0.68500000000000005
32+
; CHECK-NEXT: .section .rodata.cst8,"aM",@progbits,8
33+
; CHECK-NEXT: .p2align
34+
; CHECK-NEXT: .LCPI0_2:
35+
; CHECK-NEXT: .byte 0 // 0x0
36+
; CHECK-NEXT: .byte 4 // 0x4
37+
; CHECK-NEXT: .byte 8 // 0x8
38+
; CHECK-NEXT: .byte 12 // 0xc
39+
; CHECK-NEXT: .byte 255 // 0xff
40+
; CHECK-NEXT: .byte 255 // 0xff
41+
; CHECK-NEXT: .byte 255 // 0xff
42+
; CHECK-NEXT: .byte 255 // 0xff
43+
44+
;; Constant pools for function @unprofiled_func
45+
; CHECK: .section .rodata.cst8,"aM",@progbits,8
46+
; CHECK-NEXT: .p2align
47+
; CHECK-NEXT: .LCPI1_0:
48+
; CHECK-NEXT: .byte 0 // 0x0
49+
; CHECK-NEXT: .byte 4 // 0x4
50+
; CHECK-NEXT: .byte 8 // 0x8
51+
; CHECK-NEXT: .byte 12 // 0xc
52+
; CHECK-NEXT: .byte 255 // 0xff
53+
; CHECK-NEXT: .byte 255 // 0xff
54+
; CHECK-NEXT: .byte 255 // 0xff
55+
; CHECK-NEXT: .byte 255 // 0xff
56+
; CHECK-NEXT: .section .rodata.cst16,"aM",@progbits,16
57+
; CHECK-NEXT: .p2align
58+
; CHECK-NEXT: .LCPI1_1:
59+
; CHECK-NEXT: .word 2 // 0x2
60+
; CHECK-NEXT: .word 3 // 0x3
61+
; CHECK-NEXT: .word 5 // 0x5
62+
; CHECK-NEXT: .word 7 // 0x7
63+
; CHECK-NEXT: .section .rodata.cst16.hot,"aM",@progbits,16
64+
; CHECK-NEXT: .p2align
65+
; CHECK-NEXT: .LCPI1_2:
66+
; CHECK-NEXT: .word 442 // 0x1ba
67+
; CHECK-NEXT: .word 100 // 0x64
68+
; CHECK-NEXT: .word 0 // 0x0
69+
; CHECK-NEXT: .word 0 // 0x0
70+
71+
;; Constant pools for function @hot_func
5572
; CHECK: .section .rodata.cst8.hot,"aM",@progbits,8
56-
; CHECK: .LCPI2_0:
57-
; CHECK: .xword 0x3fe5c28f5c28f5c3 // double 0.68000000000000005
58-
; CHECK: .section .rodata.cst16.hot,"aM",@progbits,16
59-
; CHECK: .LCPI2_1:
60-
; CHECK: .word 442 // 0x1ba
61-
; CHECK: .word 100 // 0x64
62-
; CHECK: .word 0 // 0x0
63-
; CHECK: .word 0 // 0x0
64-
65-
; CHECK: .section .rodata.cst32,"aM",@progbits,32
66-
; CHECK: .globl val
73+
; CHECK-NEXT: .p2align
74+
; CHECK-NEXT: .LCPI2_0:
75+
; CHECK-NEXT: .xword 0x3fe5c28f5c28f5c3 // double 0.68000000000000005
76+
; CHECK-NEXT: .section .rodata.cst16.hot,"aM",@progbits,16
77+
; CHECK-NEXT: .p2align
78+
; CHECK-NEXT: .LCPI2_1:
79+
; CHECK-NEXT: .word 0 // 0x0
80+
; CHECK-NEXT: .word 100 // 0x64
81+
; CHECK-NEXT: .word 0 // 0x0
82+
; CHECK-NEXT: .word 442 // 0x1ba
83+
; CHECK-NEXT: .LCPI2_2:
84+
; CHECK-NEXT: .word 442 // 0x1ba
85+
; CHECK-NEXT: .word 100 // 0x64
86+
; CHECK-NEXT: .word 0 // 0x0
87+
; CHECK-NEXT: .word 0 // 0x0
88+
89+
;; For global variable @val
90+
;; The section name remains `.rodata.cst32` without hotness prefix because
91+
;; the variable has external linkage and not analyzed. Compiler need symbolized
92+
;; data access profiles to annotate such global variables' hotness.
93+
; CHECK: .section .rodata.cst32,"aM",@progbits,32
94+
; CHECK-NEXT: .globl val
6795

6896
define i32 @cold_func(double %x, <16 x i8> %a, <16 x i8> %b) !prof !16 {
6997
%2 = tail call i32 (...) @func_taking_arbitrary_param(double 6.800000e-01)
@@ -83,14 +111,16 @@ define <4 x i1> @unprofiled_func(<16 x i8> %a, <16 x i8> %b) {
83111
%t1 = call <8 x i8> @llvm.aarch64.neon.tbl2.v8i8(<16 x i8> %a, <16 x i8> %b, <8 x i8> <i8 0, i8 4, i8 8, i8 12, i8 -1, i8 -1, i8 -1, i8 -1>)
84112
%t2 = bitcast <8 x i8> %t1 to <4 x i16>
85113
%t3 = zext <4 x i16> %t2 to <4 x i32>
86-
%cmp = icmp ule <4 x i32> <i32 442, i32 100, i32 0, i32 0>, %t3
114+
%t4 = add <4 x i32> %t3, <i32 2, i32 3, i32 5, i32 7>
115+
%cmp = icmp ule <4 x i32> <i32 442, i32 100, i32 0, i32 0>, %t4
87116
ret <4 x i1> %cmp
88117
}
89118

90119
define <4 x i1> @hot_func(i32 %0, <4 x i32> %a) !prof !17 {
91120
%2 = tail call i32 (...) @func_taking_arbitrary_param(double 6.800000e-01)
92-
%b = icmp ule <4 x i32> %a, <i32 442, i32 100, i32 0, i32 0>
93-
ret <4 x i1> %b
121+
%b = add <4 x i32> <i32 0, i32 100, i32 0, i32 442>, %a
122+
%c = icmp ule <4 x i32> %b, <i32 442, i32 100, i32 0, i32 0>
123+
ret <4 x i1> %c
94124
}
95125

96126
@val = unnamed_addr constant i256 1
@@ -107,14 +137,15 @@ define i32 @main(i32 %0, ptr %1) !prof !16 {
107137

108138
7: ; preds = %7, %2
109139
%8 = phi i32 [ 0, %2 ], [ %10, %7 ]
110-
%9 = call i32 @rand()
140+
%seed_val = load i256, ptr @val
141+
%9 = call i32 @seed(i256 %seed_val)
111142
call void @hot_func(i32 %9)
112143
%10 = add i32 %8, 1
113144
%11 = icmp eq i32 %10, 100000
114145
br i1 %11, label %5, label %7, !prof !18
115146
}
116147

117-
declare i32 @rand()
148+
declare i32 @seed(i256)
118149
declare double @double_func()
119150
declare <4 x i32> @vector_func()
120151
declare <16 x i8> @vector_func_16i8()

‎llvm/test/CodeGen/X86/constant-pool-partition.ll‎

Lines changed: 49 additions & 39 deletions
Original file line numberDiff line numberDiff line change
@@ -25,42 +25,52 @@ target triple = "x86_64-grtev4-linux-gnu"
2525
; RUN: -unique-section-names=false \
2626
; RUN: %s -o - 2>&1 | FileCheck %s --dump-input=always
2727

28-
; CHECK: .section .rodata.cst8.hot,"aM",@progbits,8
29-
; CHECK: .LCPI0_0:
30-
; CHECK: .quad 0x3fe5c28f5c28f5c3 # double 0.68000000000000005
31-
; CHECK: .section .rodata.cst8.unlikely,"aM",@progbits,8
32-
; CHECK: .LCPI0_1:
33-
; CHECK: .quad 0x3eb0000000000000 # double 9.5367431640625E-7
34-
35-
; CHECK: .section .rodata.cst8,"aM",@progbits,8
36-
; CHECK: .LCPI0_2:
37-
; CHECK: .quad 0x3fc0000000000000 # double 0.125
38-
39-
; CHECK: .section .rodata.cst8,"aM",@progbits,8
40-
; CHECK: .LCPI1_0:
41-
; CHECK: .quad 0x3fc0000000000000 # double 0.125
42-
43-
; CHECK: .section .rodata.cst4,"aM",@progbits,4
44-
; CHECK: .LCPI2_0:
45-
; CHECK: .long 0x3e000000 # float 0.125
46-
47-
; CHECK: .section .rodata.cst8.hot,"aM",@progbits,8
48-
; CHECK: .LCPI3_0:
49-
; CHECK: .quad 0x3fe5c28f5c28f5c3 # double 0.68000000000000005
50-
; CHECK: .section .rodata.cst16.hot,"aM",@progbits,16
51-
; CHECK: .LCPI3_1:
52-
; CHECK: .long 2147483648 # 0x80000000
53-
; CHECK: .long 2147483648 # 0x80000000
54-
; CHECK: .long 2147483648 # 0x80000000
55-
; CHECK: .long 2147483648 # 0x80000000
56-
; CHECK: .LCPI3_2:
57-
; CHECK: .long 2147484090 # 0x800001ba
58-
; CHECK: .long 2147483748 # 0x80000064
59-
; CHECK: .long 2147483648 # 0x80000000
60-
; CHECK: .long 2147483648 # 0x80000000
61-
62-
; CHECK: .section .rodata.cst32,"aM",@progbits,32
63-
; CHECK: .globl val
28+
;; For function @cold_func
29+
; CHECK: .section .rodata.cst8.hot,"aM",@progbits,8
30+
; CHECK-NEXT: .p2align
31+
; CHECK-NEXT: .LCPI0_0:
32+
; CHECK-NEXT: .quad 0x3fe5c28f5c28f5c3 # double 0.68000000000000005
33+
; CHECK-NEXT: .section .rodata.cst8.unlikely,"aM",@progbits,8
34+
; CHECK-NEXT: .p2align
35+
; CHECK-NEXT: .LCPI0_1:
36+
; CHECK-NEXT: .quad 0x3eb0000000000000 # double 9.5367431640625E-7
37+
; CHECK-NEXT: .section .rodata.cst8,"aM",@progbits,8
38+
; CHECK-NEXT: .p2align
39+
; CHECK-NEXT: .LCPI0_2:
40+
; CHECK-NEXT: .quad 0x3fc0000000000000 # double 0.125
41+
42+
;; For function @unprofiled_func_double
43+
; CHECK: .section .rodata.cst8,"aM",@progbits,8
44+
; CHECK-NEXT: .p2align
45+
; CHECK-NEXT: .LCPI1_0:
46+
; CHECK-NEXT: .quad 0x3fc0000000000000 # double 0.125
47+
48+
;; For function @unprofiled_func_float
49+
; CHECK: .section .rodata.cst4,"aM",@progbits,4
50+
; CHECK-NEXT: .p2align
51+
; CHECK-NEXT: .LCPI2_0:
52+
; CHECK-NEXT: .long 0x3e000000 # float 0.125
53+
54+
;; For function @hot_func
55+
; CHECK: .section .rodata.cst8.hot,"aM",@progbits,8
56+
; CHECK-NEXT: .p2align
57+
; CHECK-NEXT: .LCPI3_0:
58+
; CHECK-NEXT: .quad 0x3fe5c28f5c28f5c3 # double 0.68000000000000005
59+
; CHECK-NEXT: .section .rodata.cst16.hot,"aM",@progbits,16
60+
; CHECK-NEXT: .p2align
61+
; CHECK-NEXT: .LCPI3_1:
62+
; CHECK-NEXT: .long 2147483648 # 0x80000000
63+
; CHECK-NEXT: .long 2147483648 # 0x80000000
64+
; CHECK-NEXT: .long 2147483648 # 0x80000000
65+
; CHECK-NEXT: .long 2147483648 # 0x80000000
66+
; CHECK-NEXT: .LCPI3_2:
67+
; CHECK-NEXT: .long 2147484090 # 0x800001ba
68+
; CHECK-NEXT: .long 2147483748 # 0x80000064
69+
; CHECK-NEXT: .long 2147483648 # 0x80000000
70+
; CHECK-NEXT: .long 2147483648 # 0x80000000
71+
72+
; CHECK: .section .rodata.cst32,"aM",@progbits,32
73+
; CHECK-NEXT: .globl val
6474

6575
define double @cold_func(double %x) !prof !16 {
6676
%2 = tail call i32 (...) @func_taking_arbitrary_param(double 6.800000e-01)
@@ -79,7 +89,6 @@ define float @unprofiled_func_float(float %x) {
7989
ret float %z
8090
}
8191

82-
8392
define <4 x i1> @hot_func(i32 %0, <4 x i32> %a) !prof !17 {
8493
%2 = tail call i32 (...) @func_taking_arbitrary_param(double 6.800000e-01)
8594
%b = icmp ule <4 x i32> %a, <i32 442, i32 100, i32 0, i32 0>
@@ -98,14 +107,15 @@ define i32 @main(i32 %0, ptr %1) !prof !16 {
98107

99108
7: ; preds = %7, %2
100109
%8 = phi i32 [ 0, %2 ], [ %10, %7 ]
101-
%9 = call i32 @rand()
110+
%seed_val = load i256, ptr @val
111+
%9 = call i32 @seed(i256 %seed_val)
102112
call void @hot_func(i32 %9)
103113
%10 = add i32 %8, 1
104114
%11 = icmp eq i32 %10, 100000
105115
br i1 %11, label %5, label %7, !prof !18
106116
}
107117

108-
declare i32 @rand()
118+
declare i32 @seed(i256)
109119
declare double @double_func()
110120
declare i32 @func_taking_arbitrary_param(...)
111121

0 commit comments

Comments
 (0)