1 /* { dg-do run { target { powerpc*-*-* && { lp64 && p9vector_hw } } } } */
2 /* { dg-options "-mdejagnu-cpu=power9 -O2 " } */
3
4 /* PR83677: This test case used to fail due to mis-generation of the
5 xxpermr instruction. It requires inlining to create enough register
6 pressure that we generate xxpermr rather than vpermr. */
7
8 #include <altivec.h>
9
v_expand_u8(vector unsigned char * a,vector unsigned short * b0,vector unsigned short * b1)10 void v_expand_u8(vector unsigned char* a, vector unsigned short* b0, vector unsigned short* b1)
11 {
12 *b0 = (vector unsigned short)vec_mergeh(*a, vec_splats((unsigned char)0));
13 *b1 = (vector unsigned short)vec_mergel(*a, vec_splats((unsigned char)0));
14 }
15
v_expand_u16(vector unsigned short * a,vector unsigned int * b0,vector unsigned int * b1)16 void v_expand_u16(vector unsigned short* a, vector unsigned int* b0, vector unsigned int* b1)
17 {
18 *b0 = (vector unsigned int)vec_mergeh(*a, vec_splats((unsigned short)0));
19 *b1 = (vector unsigned int)vec_mergel(*a, vec_splats((unsigned short)0));
20 }
21
v_load_deinterleave_u8(unsigned char * ptr,vector unsigned char * a,vector unsigned char * b,vector unsigned char * c)22 void v_load_deinterleave_u8(unsigned char *ptr, vector unsigned char* a, vector unsigned char* b, vector unsigned char* c)
23 {
24 vector unsigned char v1 = vec_xl( 0, ptr);
25 vector unsigned char v2 = vec_xl(16, ptr);
26 vector unsigned char v3 = vec_xl(32, ptr);
27
28 static const vector unsigned char a12_perm = {0, 3, 6, 9, 12, 15, 18, 21, 24, 27, 30, 0, 0, 0, 0, 0};
29 static const vector unsigned char a123_perm = {0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 17, 20, 23, 26, 29};
30 *a = vec_perm(vec_perm(v1, v2, a12_perm), v3, a123_perm);
31
32 static const vector unsigned char b12_perm = {1, 4, 7, 10, 13, 16, 19, 22, 25, 28, 31, 0, 0, 0, 0, 0};
33 static const vector unsigned char b123_perm = {0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 18, 21, 24, 27, 30};
34 *b = vec_perm(vec_perm(v1, v2, b12_perm), v3, b123_perm);
35
36 static const vector unsigned char c12_perm = {2, 5, 8, 11, 14, 17, 20, 23, 26, 29, 0, 0, 0, 0, 0, 0};
37 static const vector unsigned char c123_perm = {0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 16, 19, 22, 25, 28, 31};
38 *c = vec_perm(vec_perm(v1, v2, c12_perm), v3, c123_perm);
39 }
40
v_load_deinterleave_f32(float * ptr,vector float * a,vector float * b,vector float * c)41 void v_load_deinterleave_f32(float *ptr, vector float* a, vector float* b, vector float* c)
42 {
43 vector float v1 = vec_xl( 0, ptr);
44 vector float v2 = vec_xl(16, ptr);
45 vector float v3 = vec_xl(32, ptr);
46
47 static const vector unsigned char flp = {0, 1, 2, 3, 12, 13, 14, 15, 16, 17, 18, 19, 28, 29, 30, 31};
48 *a = vec_perm(v1, vec_sld(v3, v2, 8), flp);
49
50 static const vector unsigned char flp2 = {28, 29, 30, 31, 0, 1, 2, 3, 12, 13, 14, 15, 16, 17, 18, 19};
51 *b = vec_perm(v2, vec_sld(v1, v3, 8), flp2);
52
53 *c = vec_perm(vec_sld(v2, v1, 8), v3, flp);
54 }
55
v_store_interleave_f32(float * ptr,vector float a,vector float b,vector float c)56 void v_store_interleave_f32(float *ptr, vector float a, vector float b, vector float c)
57 {
58 vector float hbc = vec_mergeh(b, c);
59
60 static const vector unsigned char ahbc = {0, 1, 2, 3, 16, 17, 18, 19, 20, 21, 22, 23, 4, 5, 6, 7};
61 vec_xst(vec_perm(a, hbc, ahbc), 0, ptr);
62
63 vector float lab = vec_mergel(a, b);
64 vec_xst(vec_sld(lab, hbc, 8), 16, ptr);
65
66 static const vector unsigned char clab = {8, 9, 10, 11, 24, 25, 26, 27, 28, 29, 30, 31, 12, 13, 14, 15};
67 vec_xst(vec_perm(c, lab, clab), 32, ptr);
68 }
69
v_cvt_f32(vector unsigned int a)70 vector float v_cvt_f32(vector unsigned int a)
71 {
72 return (vector float)vec_ctf(a, 0);
73 }
74
acc_simd_(const unsigned char * src,float * dst,const unsigned char * mask,int len)75 void acc_simd_(const unsigned char* src, float* dst, const unsigned char* mask, int len)
76 {
77 int x = 0;
78 const int cVectorWidth = 16;
79
80 for ( ; x <= len - cVectorWidth; x += cVectorWidth)
81 {
82 vector unsigned char v_mask = vec_xl(0, mask + x);
83 v_mask = (vector unsigned char)vec_cmpeq(vec_splats((unsigned char)0), v_mask);
84 v_mask = (vector unsigned char)vec_nor(v_mask, v_mask);
85 vector unsigned char v_src0, v_src1, v_src2;
86 v_load_deinterleave_u8((unsigned char *)(src + (x * 3)), &v_src0, &v_src1, &v_src2);
87 v_src0 = v_src0 & v_mask;
88 v_src1 = v_src1 & v_mask;
89 v_src2 = v_src2 & v_mask;
90
91 /* expand 16 uchar to 4 vectors which contains 4 uint */
92 vector unsigned short v_src00, v_src01, v_src10, v_src11, v_src20, v_src21;
93 v_expand_u8(&v_src0, &v_src00, &v_src01);
94 v_expand_u8(&v_src1, &v_src10, &v_src11);
95 v_expand_u8(&v_src2, &v_src20, &v_src21);
96 vector unsigned int v_src000, v_src001, v_src010, v_src011;
97 vector unsigned int v_src100, v_src101, v_src110, v_src111;
98 vector unsigned int v_src200, v_src201, v_src210, v_src211;
99 v_expand_u16(&v_src00, &v_src000, &v_src001);
100 v_expand_u16(&v_src01, &v_src010, &v_src011);
101 v_expand_u16(&v_src10, &v_src100, &v_src101);
102 v_expand_u16(&v_src11, &v_src110, &v_src111);
103 v_expand_u16(&v_src20, &v_src200, &v_src201);
104 v_expand_u16(&v_src21, &v_src210, &v_src211);
105
106 vector float v_dst000, v_dst001, v_dst010, v_dst011;
107 vector float v_dst100, v_dst101, v_dst110, v_dst111;
108 vector float v_dst200, v_dst201, v_dst210, v_dst211;
109 v_load_deinterleave_f32(dst + (x * 3), &v_dst000, &v_dst100, &v_dst200);
110 v_load_deinterleave_f32(dst + ((x + 4) * 3), &v_dst001, &v_dst101, &v_dst201);
111 v_load_deinterleave_f32(dst + ((x + 8) * 3), &v_dst010, &v_dst110, &v_dst210);
112 v_load_deinterleave_f32(dst + ((x + 12) * 3), &v_dst011, &v_dst111, &v_dst211);
113
114 v_store_interleave_f32(dst + (x * 3), vec_add(v_dst000, v_cvt_f32(v_src000)), vec_add(v_dst100, v_cvt_f32(v_src100)), vec_add(v_dst200, v_cvt_f32(v_src200)));
115 v_store_interleave_f32(dst + ((x + 4) * 3), vec_add(v_dst001, v_cvt_f32(v_src001)), vec_add(v_dst101, v_cvt_f32(v_src101)), vec_add(v_dst201, v_cvt_f32(v_src201)));
116 v_store_interleave_f32(dst + ((x + 8) * 3), vec_add(v_dst010, v_cvt_f32(v_src010)), vec_add(v_dst110, v_cvt_f32(v_src110)), vec_add(v_dst210, v_cvt_f32(v_src210)));
117 v_store_interleave_f32(dst + ((x + 12) * 3), vec_add(v_dst011, v_cvt_f32(v_src011)), vec_add(v_dst111, v_cvt_f32(v_src111)), vec_add(v_dst211, v_cvt_f32(v_src211)));
118 }
119 return;
120 }
121
acc_(const unsigned char * src,float * dst,const unsigned char * mask,int len)122 void acc_(const unsigned char* src, float* dst, const unsigned char* mask, int len)
123 {
124 int x = 0;
125 src += (x * 3);
126 dst += (x * 3);
127 for( ; x < len; x++, src += 3, dst += 3 )
128 {
129 if( mask[x] ) /* if mask, R/G/B dst[] += src[] */
130 {
131 for( int k = 0; k < 3; k++ )
132 {
133 dst[k] += src[k];
134 }
135 }
136 }
137 return;
138 }
139
140 #define N 16
141
main(int argc,char * argv[])142 int main(int argc, char *argv[])
143 {
144 unsigned char __attribute__ ((aligned (16) )) mask[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 1, 0, 0, 1};
145 unsigned char __attribute__ ((aligned (16) )) src[3*N];
146 float __attribute__ ((aligned (16) )) dst[3*N];
147 float __attribute__ ((aligned (16) )) exp[3*N];
148
149 int i;
150
151 /* initialize src and dst */
152 for (i=0; i<3*N; i++) src[i] = (unsigned char)(i*3);
153 for (i=0; i<3*N; i++) {dst[i] = i * 1.0f; exp[i] = dst[i];}
154
155 acc_(src, exp, mask, N);
156 acc_simd_(src, dst, mask, N);
157
158 for (i=0; i<N; i++)
159 {
160 if ((dst[3*i] != exp[3*i]) || (dst[3*i+1] != exp[3*i+1]) || (dst[3*i+2] != exp[3*i+2]))
161 __builtin_abort ();
162 }
163
164 return 0;
165 }
166