1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
|
From 88516507757932c1e67ce99d240596935971d2d0 Mon Sep 17 00:00:00 2001
From: liuhongt <hongtao.liu@intel.com>
Date: Thu, 9 Nov 2023 13:20:05 +0800
Subject: [PATCH 13/28] Fix wrong code due to vec_merge + pcmp to blendvb
splitter.
gcc/ChangeLog:
PR target/112443
* config/i386/sse.md (*avx2_pcmp<mode>3_4): Fix swap condition
from LT to GT since there's not in the pattern.
(*avx2_pcmp<mode>3_5): Ditto.
gcc/testsuite/ChangeLog:
* g++.target/i386/pr112443.C: New test.
(cherry picked from commit 9a0cc04b9c9b02426762892b88efc5c44ba546bd)
---
gcc/config/i386/sse.md | 4 +-
gcc/testsuite/g++.target/i386/pr112443.C | 108 +++++++++++++++++++++++
2 files changed, 110 insertions(+), 2 deletions(-)
create mode 100644 gcc/testsuite/g++.target/i386/pr112443.C
diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md
index f25dd5f2b..23b858ab2 100644
--- a/gcc/config/i386/sse.md
+++ b/gcc/config/i386/sse.md
@@ -16358,7 +16358,7 @@
(match_dup 4))]
UNSPEC_BLENDV))]
{
- if (INTVAL (operands[5]) == 1)
+ if (INTVAL (operands[5]) == 5)
std::swap (operands[1], operands[2]);
operands[3] = gen_lowpart (<MODE>mode, operands[3]);
})
@@ -16388,7 +16388,7 @@
(match_dup 4))]
UNSPEC_BLENDV))]
{
- if (INTVAL (operands[5]) == 1)
+ if (INTVAL (operands[5]) == 5)
std::swap (operands[1], operands[2]);
})
diff --git a/gcc/testsuite/g++.target/i386/pr112443.C b/gcc/testsuite/g++.target/i386/pr112443.C
new file mode 100644
index 000000000..ebfa9b4a7
--- /dev/null
+++ b/gcc/testsuite/g++.target/i386/pr112443.C
@@ -0,0 +1,108 @@
+/* { dg-do run } */
+/* { dg-require-effective-target avx512bw } */
+/* { dg-require-effective-target avx512vl } */
+/* { dg-options "-O2 -std=c++17 -mavx512bw -mavx512vl" } */
+
+#include <cstdint>
+#include <x86intrin.h>
+#include <functional>
+#include <ostream>
+
+#define AVX512BW
+#define AVX512VL
+
+#include "avx512f-helper.h"
+
+struct TensorIteratorBase{
+ char* in;
+ char* out;
+
+ void for_each(std::function<void(char*, char*, int64_t size)> loop){
+ loop(out, in, 32);
+ }
+};
+
+class Vectorized {
+protected:
+ __m256i values;
+
+ static inline __m256i invert(const __m256i& v) {
+ const auto ones = _mm256_set1_epi64x(-1);
+ return _mm256_xor_si256(ones, v);
+ }
+public:
+ operator __m256i() const {
+ return values;
+ }
+
+ static constexpr int size() {
+ return 32;
+ }
+
+ Vectorized() {}
+ Vectorized(__m256i v) : values(v) {}
+ Vectorized(uint8_t v) { values = _mm256_set1_epi8(v); }
+ static Vectorized blendv(const Vectorized& a, const Vectorized& b,
+ const Vectorized& mask) {
+ return _mm256_blendv_epi8(a, b, mask);
+ }
+ static Vectorized loadu(const void* ptr) {
+ return _mm256_loadu_si256(reinterpret_cast<const __m256i*>(ptr));
+ }
+ void store(void* ptr) const {
+ _mm256_storeu_si256(reinterpret_cast<__m256i*>(ptr), values);
+ }
+
+ Vectorized operator<(const Vectorized& other) const {
+ __m256i max = _mm256_max_epu8(values, other);
+ return invert(_mm256_cmpeq_epi8(max, values));
+ }
+ Vectorized operator-(const Vectorized& b) {
+ return _mm256_sub_epi8(values, b);
+ }
+};
+
+std::ostream& operator<<(std::ostream& stream, const Vectorized& vec) {
+ uint8_t buf[Vectorized::size()];
+ vec.store(buf);
+ stream << "vec[";
+ for (int i = 0; i != Vectorized::size(); i++) {
+ if (i != 0)
+ stream << ", ";
+ stream << buf[i]*1;
+ }
+ stream << "]";
+ return stream;
+}
+
+void run(TensorIteratorBase iter){
+ Vectorized zero_vec(0);
+ Vectorized one_vec(1);
+
+ iter.for_each([=](char* out, char* in, int64_t size) {
+ for (int64_t i = 0; i <= size - Vectorized::size(); i += Vectorized::size()) {
+ auto self_vec = Vectorized::loadu(in + i);
+ auto left = Vectorized::blendv(zero_vec, one_vec, zero_vec < self_vec);
+ auto right = Vectorized::blendv(zero_vec, one_vec, self_vec < zero_vec);
+ auto outv = left - right;
+ outv.store(out + i);
+ }
+ });
+}
+
+void
+test_256 (){
+ char in[32];
+ char out[32];
+ for(auto& x: in) x = 1;
+ run(TensorIteratorBase{in, out});
+ Vectorized::loadu (out);
+ for (int i = 0; i != 32; i++)
+ if (out[i] != 1)
+ __builtin_abort ();
+}
+
+void
+test_128 ()
+{
+}
--
2.31.1
|
