代码

// ===== 引用区（本发增量）=====
// [1] duck.ac 用户 saffah_v41f1017_agg4（本账号），提交 #135851
//     <https://duck.ac/submission/135851>（1,824.830021 ms） —— 本发底盘（正文逐字节副本；当前队内 best=#135851 ordB 1,824.830021）。
//     本发仅动：ordB 旋转 + kpA 前 .space128 组合（新 .text 类 265c3f8b，196,827B）（单变量/组合；值精确、.text 唯一）。
// LICENSE：duck.ac 公开提交；无独立许可证声明；按「账号+URL+沿用内容」具名引用。
// ===== 思路区（本发）=====
// 【组合·序+垫】ordB（case 旋转 [7,0..6]，+135851）再于 kpA 前加 .space 128；序类与 128B 位移叠加，语义/值不变。
// 【闸门】本地 n=4096 双趟 driver FNV=57b2ecc0f4cfdb44 与底盘逐位同；无 main；chars≤102,400。
// ==== 以下为底盘原文（逐字节保留）====
// ===== 引用区（本发增量）=====
// [1] 本账号 saffah_v41f1017_agg4，提交 #134764 <https://duck.ac/submission/134764>
//     （题 mmmd4k，1,826.850554 ms，rank1）—— 本发底盘；除本发下述单变量外正文逐字节保留（其完整引用链/思路区保留于下）。
// [2] duck.ac 题目页 <https://duck.ac/problem/mmmd4k>（站点公开条款：提交正文公开可见）。
// LICENSE：以上均为 duck.ac 公开提交，无独立许可证声明、非来自任何外部仓库；按「账号+URL+沿用内容」具名引用，
//     未声称引用部分为原创。本发未参考任何已发表论文 / 外部开源仓库。
// ===== 思路区（本发）=====
// 【单变量】D 相位三条冷载（L0/L2/L1）背靠背先发、再算 D1/D0（kpA/kpB 各 1 处；三临时 ymm13）。值中性。依据：in-engine 曾测 −0.08%（噪声内）、未官方定价；官方 i3 内存域不同 ⇒ 官方单发定价。
// 【闸门】本地 driver n=4096 chk=57b2ecc0f4cfdb44 与底盘逐位同；判题同款 g++-9 .text md5=acb466a001bd56efefd7abaa8343eec6 ≠ 底盘 573ad7e1…（及他席提交件）。
// 【口径】官方 i3 内存域与 worker 不同（3.6× gather 已证）⇒ 官方唯一通道单发（空窗、Pending≤1）；无 main。
// ===== 以下为 #134764 原文（逐字节保留）=====
/* ===== 引用区（本发新增段；以下原有引用块原样保留）=====
 * [1] duck.ac 用户 saffah_v41f1017_agg4（本账号），提交 #134745
 *     <https://duck.ac/submission/134745>（`mmmd4k`，1.827132803 s，现役最好）—— 本件正文 = 该提交正文逐字节副本；
 *     本发仅改 root_prefetch（h=2048 顶层合并 6 流 T1 提示）的目标位置：沿根序领先 (c11,c21,c22,m,m5,m6)
 *     = (0,0,0,32,32,32) 个 double（0=现役“下一 root_step 块”）。上游 Credit 链原样保留。
 * LICENSE：duck.ac 公开提交；无独立许可证声明 ⇒ 按具名引用使用。
 * ===== 思路区（本发）=====
 * 【单变量·root 预取逐流距离 {tag}】仅 m 三平面流 +32（C 三流保持 0）——组分解。
 *   地址按 (pos+d) 重算 (i,j)（32 的倍数 ⇒ 行内不跨界；C 流跨行处仍精确指向 pos+d 的真实块）。
 *   纯提示、值逐位不变。
 * 【闸门】本地 driver n=4096 chk=57b2ecc0f4cfdb44 与基座逐位同；-c 通过。
 * ============================================================ */
/* ===== 引用区（本发新增段；以下原有引用块原样保留）=====
 * [1] duck.ac 用户 saffah_v41f1017_agg4（本账号），提交 #134596
 *     <https://duck.ac/submission/134596>（`mmmd4k`，1.828935163 s）——本件正文 = 该提交正文逐字节副本；
 *     本发仅在其 kpA/kpB 手写 asm 的 D 相位前插入 3 条 prefetcht0（逐流距离 L0=32B/L1=64B/L2=96B）。上游 Credit 链原样保留。
 * LICENSE：duck.ac 公开提交；无独立许可证声明 ⇒ 按具名引用使用。
 * ===== 思路区（本发）=====
 * 【单变量·D 相位逐流预取距离：L0=32B / L1=64B / L2=96B】逐流官方单发矩阵（vs 全 32=1,827.625481）：
 *   L0=64 −0.183、L1=64 −0.235、L2=64 −0.399；配对 L0L1 −0.201、L0L2 −0.343、**L1L2 −0.436（best）**；全 64 +0.153
 *   ⇒ 三流共享资源反协同。本发检验主导流 L2 的更深距离（96B）。 纯提示、值逐位不变。
 * 【闸门】本地 driver n=4096 chk=57b2ecc0f4cfdb44 与基座逐位同；-c 通过。
 * ============================================================ */
/* ===== 引用区 =====
 * [1] duck.ac 用户 saffah_v41f1017_agg4（本账号），提交 #133582 <https://duck.ac/submission/133582>
 *     （`mmmd4k`，1.828939701 s，现役最好/rank1）—— 本件正文的直接基座（含其自带整条上游
 *     Credit 链与许可说明，逐字节保留）；本发相对基座仅改 .text 头部死填充字节数（见思路区）。
 * LICENSE：duck.ac 公开提交；无独立许可证声明、非来自任何外部仓库 ⇒ 按具名引用使用。
 * ===== 思路区 =====
 * 【本发单机制·细 D 放置格】基座为 D=0+AF128；既有 D 图是 128B 步长粗扫（D0 最优，+128B +0.30ms、
 *   +256B +0.76ms），而姊妹题 b4k 刚证明 8B 步长细网格可在 D0 之外找到谷底。本发为 AF128 固定下的
 *   细 D 单变量格：D=72 字节（__asm__(".space 72,0x90")，纯死填充、不执行）。
 * 【闸门】n=4096 seed999 全 C dump md5 与基座逐位相同（d4a8181820b2b571d7a7c81965cd43fd）。
 * 【目的】判题机 Time (ms) ns 级通道定价放置格；不改任何算法/常数/循环。
 * ============================================================ */

/* ===== 引用区 =====
 * [1] duck.ac 用户 saffah_v41f1017_agg4（本账号），提交 #133057 <https://duck.ac/submission/133057>
 *     （`mmmd4k`，1.830 s）—— **本件正文的直接基座**：该件正文与其自带的整条上游 Credit 链
 *     （saffah_cc_v41_agg1 #127899 及更早各代，含许可说明）逐字节保留；本发相对基座仅加
 *     [SCRATCH-CHAIN] 一处写侧改动（见思路区）。
 * [2] duck.ac 用户 saffah_v41f1017_agg4（本账号），提交 #133177 与 #133241
 *     <https://duck.ac/submission/133177> <https://duck.ac/submission/133241>（`mmml4k`，
 *     n=4096 姊妹引擎，已达标）—— **沿用其写侧机理**「NT 流存阈值 h≥512」：对 h≥512
 *     （缓冲 > L3）的变换目的地用 vmovntpd/vmovntdq 流存，消除 WB 写的 RFO（该题板面 −33ms）。
 *     两引擎不同构，本件未复制其代码，仅按机理在 d4k pre 传送带的剩余 WB 站点实现。
 * LICENSE：以上均为 duck.ac 公开提交（站点 FAQ 明示提交代码公开可见）；无第三方许可证声明、
 *     非来自任何外部开源仓库 ⇒ 按具名引用（账号 + 提交地址 + 沿用内容/机理）使用。
 *     本件未参考任何已发表论文或外部开源仓库。
 * ===== 思路区 =====
 * 【本发单机制·写侧 RFO 消除】d4k pre 传送带每侧 4 个和面板中已有 3 个为 NT 流存；唯独
 *   pass0 的 p 面板必须 WB，因为 pass1 要在下一次 chunk 读它（串行链）。本发把 p 面板也改
 *   NT（8.4/33.5MB >> L3，消费者在子3之后才读，冷），同时把同一 slab 另写一份到按任务槽位
 *   索引的 2KB L1 scratch（d4ah_scr[8][256]），pass1 的链式操作数改指该 scratch。
 *   ⇒ 传送带写入全部 NT，冷写 RFO 全消除；DRAM 口径 5R+4W → 4R+4W
 *   （根级省 67MB + 子层省 118MB = 185MB 读/跑）。
 * 【值】逐位不变：scratch 存的就是原先写进 p 面板的同一 fl 值；pass1 的链式位置按侧别替换
 *   （side0 替换 L0、side1 替换 L1）。
 * 【闸门】n=4096 种子 999/4242 全 C FNV 与基座逐位同（47072e456fffc0a4 / a60bb5e71467f7e7）；
 *   本地 A/B 中性（VM 噪声 ±1% 内）。
 * 【目的（试验性提交须写）】判题机板面为唯一判据：验证 mmml4k 的 NT512 写侧收益在同尺寸
 *   d4k 引擎最后一个冷 WB 站点上是否兑现。
 * ============================================================ */

#include <cstdint>
#pragma GCC optimize("O3,unroll-loops,no-strict-aliasing")
#pragma GCC optimize("tracer")
#pragma GCC optimize("align-functions=128")
#line 309
#include <sys/mman.h>
#pragma GCC target("avx2,fma,arch=broadwell")
#include <immintrin.h>
__asm__(".space 72,0x90");
#ifndef B2M_NO_BARA
#define B2M_NO_BARA 1
#endif
#define MD9_T2BASE 1
#include <stddef.h>
#ifndef BASE
#define BASE 512
#endif
#ifndef MC
#define MC 30
#endif
#ifndef NTMIN
#define NTMIN 512
#endif
#ifndef UPPER_PAGE_PAD
#define UPPER_PAGE_PAD 15872
#endif
#ifndef LEAF_PAD
#define LEAF_PAD 1152
#endif
#ifndef SUMNT
#define SUMNT 2048
#endif
static constexpr int MR=6,NR=8;
alignas(4096) static double workspace[48*1024*1024];
alignas(2048) static double ap[(MC+MR)*(BASE+24)];
alignas(1024) static double ap2[(MC+MR)*(BASE+24)];
alignas(64) static double bp[BASE*(BASE+24)];
alignas(64) static double pk_zero[BASE+8];
alignas(64) static double pk_mp[4]={0.0,0.0,0.0,0.0}, pk_mn[4]={-0.0,-0.0,-0.0,-0.0};
struct Mat { const double *x; int ld; const double *y=nullptr; int ly=0; int sign=0; };
static inline Mat d4bd_first(Mat A,int S){ A.x=A.x+(S-2*BASE); A.y=A.x+BASE; A.ly=A.ld; A.sign=1; return A; }
static int d4bd_np=0;
static inline __m256d ld(Mat a,int i,int j) {
__m256d x=_mm256_load_pd(a.x+(size_t)i*a.ld+j);
if(a.sign) {
__m256d y=_mm256_load_pd(a.y+(size_t)i*a.ly+j);
x=a.sign>0?_mm256_add_pd(x,y):_mm256_sub_pd(x,y);
}
return x;
}
static inline double scalar(Mat a,int i,int k) {
double x=a.x[(size_t)i*a.ld+k];
if(a.sign)x+=a.sign>0?a.y[(size_t)i*a.ly+k]:-a.y[(size_t)i*a.ly+k];
return x;
}
#define D4SRC_FMA(K,P) "vbroadcastsd %c[o0]+" #K "(%[a],%[i]), %%ymm14\n\t" \
"vfmadd231pd %%ymm12, %%ymm14, %%ymm0\n\t" \
"vfmadd231pd %%ymm13, %%ymm14, %%ymm1\n\t" \
"vbroadcastsd %c[o1]+" #K "(%[a],%[i]), %%ymm14\n\t" \
"vfmadd231pd %%ymm12, %%ymm14, %%ymm2\n\t" \
"vfmadd231pd %%ymm13, %%ymm14, %%ymm3\n\t" \
"prefetcht0 " #P "(%[b],%[i],8)\n\t" \
"vbroadcastsd %c[o2]+" #K "(%[a],%[i]), %%ymm14\n\t" \
"vfmadd231pd %%ymm12, %%ymm14, %%ymm4\n\t" \
"vfmadd231pd %%ymm13, %%ymm14, %%ymm5\n\t" \
"vbroadcastsd %c[o3]+" #K "(%[a],%[i]), %%ymm14\n\t" \
"vfmadd231pd %%ymm12, %%ymm14, %%ymm6\n\t" \
"vfmadd231pd %%ymm13, %%ymm14, %%ymm7\n\t" \
"vbroadcastsd %c[o4]+" #K "(%[a],%[i]), %%ymm14\n\t" \
"vfmadd231pd %%ymm12, %%ymm14, %%ymm8\n\t" \
"vfmadd231pd %%ymm13, %%ymm14, %%ymm9\n\t" \
"vbroadcastsd %c[o5]+" #K "(%[a],%[i]), %%ymm14\n\t" \
"vfmadd231pd %%ymm12, %%ymm14, %%ymm10\n\t" \
"vfmadd231pd %%ymm13, %%ymm14, %%ymm11\n\t"
template<int SG> static __attribute__((always_inline)) inline void kernel(int n,const double *a,const double *b,double *c,int ldc,int rows,
const double *pks,const double *pky,double *pkd,int source_ld,int source_ly) {
double *out=c;
ptrdiff_t stride=(ptrdiff_t)ldc*8;
long k=-(long)n*4; long ip=-(long)(n/8)*64;
pks+=n/2;pky+=n/2;pkd+=n/2;static_assert(SG!=0,"indexed carried packing");
const char *pfs=(const char*)((uintptr_t)pks+source_ld*8+128),*pfy=(const char*)((uintptr_t)pky+source_ly*8);
asm volatile (
"vxorpd %%ymm0, %%ymm0, %%ymm0\n\t"
"vxorpd %%ymm1, %%ymm1, %%ymm1\n\t"
"vxorpd %%ymm2, %%ymm2, %%ymm2\n\t"
"vxorpd %%ymm3, %%ymm3, %%ymm3\n\t"
"vxorpd %%ymm4, %%ymm4, %%ymm4\n\t"
"vxorpd %%ymm5, %%ymm5, %%ymm5\n\t"
"vxorpd %%ymm6, %%ymm6, %%ymm6\n\t"
"vxorpd %%ymm7, %%ymm7, %%ymm7\n\t"
"vxorpd %%ymm8, %%ymm8, %%ymm8\n\t"
"vxorpd %%ymm9, %%ymm9, %%ymm9\n\t"
"vxorpd %%ymm10, %%ymm10, %%ymm10\n\t"
"vxorpd %%ymm11, %%ymm11, %%ymm11\n\t"
".p2align 6\n\t"
"1:\n\t"
"prefetcht0 (%[pfs],%[k])\n\t"
".if %c[sg] == 1 || %c[sg] == -1\n\t"
"prefetcht0 (%[pfy],%[k])\n\t"
".endif\n\t"
"vmovapd 32768(%[b],%[i],8), %%ymm12\n\t"
"vmovapd 32800(%[b],%[i],8), %%ymm13\n\t"
D4SRC_FMA(4096,33280)
"vmovapd 32832(%[b],%[i],8), %%ymm12\n\t"
"vmovapd 32864(%[b],%[i],8), %%ymm13\n\t"
D4SRC_FMA(4104,33344)
"vmovupd 0(%[pks],%[k]), %%ymm15\n\t"
"vmovapd 32896(%[b],%[i],8), %%ymm12\n\t"
"vmovapd 32928(%[b],%[i],8), %%ymm13\n\t"
D4SRC_FMA(4112,33408)
"vmovapd 32960(%[b],%[i],8), %%ymm12\n\t"
"vmovapd 32992(%[b],%[i],8), %%ymm13\n\t"
D4SRC_FMA(4120,33472)
".if %c[sg] == 1\n\t"
"vaddpd 0(%[pky],%[k]), %%ymm15, %%ymm15\n\t"
".elseif %c[sg] == -1\n\t"
"vsubpd 0(%[pky],%[k]), %%ymm15, %%ymm15\n\t"
".endif\n\t"
"vmovapd 33024(%[b],%[i],8), %%ymm12\n\t"
"vmovapd 33056(%[b],%[i],8), %%ymm13\n\t"
D4SRC_FMA(4128,33536)
"vmovapd 33088(%[b],%[i],8), %%ymm12\n\t"
"vmovapd 33120(%[b],%[i],8), %%ymm13\n\t"
D4SRC_FMA(4136,33600)
"vmovapd 33152(%[b],%[i],8), %%ymm12\n\t"
"vmovapd 33184(%[b],%[i],8), %%ymm13\n\t"
D4SRC_FMA(4144,33664)
"vmovapd 33216(%[b],%[i],8), %%ymm12\n\t"
"vmovapd 33248(%[b],%[i],8), %%ymm13\n\t"
D4SRC_FMA(4152,33728)
"vmovupd %%ymm15, 0(%[pkd],%[k])\n\t"
"add $32, %[k]\n\t"
"add $64, %[i]\n\t"
"jnz 1b\n\t"
"vmovntpd %%ymm0, (%[out])\n\t"
"vmovntpd %%ymm1, 32(%[out])\n\t"
"vmovntpd %%ymm2, (%[out],%[stride])\n\t"
"vmovntpd %%ymm3, 32(%[out],%[stride])\n\t"
"vmovntpd %%ymm4, (%[out],%[stride],2)\n\t"
"vmovntpd %%ymm5, 32(%[out],%[stride],2)\n\t"
"lea (%[out],%[stride],2), %[out]\n\t"
"add %[stride], %[out]\n\t"
"vmovntpd %%ymm6, (%[out])\n\t"
"vmovntpd %%ymm7, 32(%[out])\n\t"
"vmovntpd %%ymm8, (%[out],%[stride])\n\t"
"vmovntpd %%ymm9, 32(%[out],%[stride])\n\t"
"vmovntpd %%ymm10, (%[out],%[stride],2)\n\t"
"vmovntpd %%ymm11, 32(%[out],%[stride],2)\n\t"
: [a] "+&r"(a), [b] "+&r"(b), [out] "+&r"(out), [k] "+&r"(k), [i] "+&r"(ip),
[pks] "+&r"(pks), [pky] "+&r"(pky), [pkd] "+&r"(pkd)
: [pfs] "r"(pfs), [pfy] "r"(pfy), [stride] "r"(stride), [sg] "i"(SG), [o0] "i"(0*(BASE+24)*8), [o1] "i"(1*(BASE+24)*8), [o2] "i"(2*(BASE+24)*8), [o3] "i"(3*(BASE+24)*8), [o4] "i"(4*(BASE+24)*8), [o5] "i"(5*(BASE+24)*8)
: "memory", "cc", "ymm0","ymm1","ymm2","ymm3","ymm4","ymm5","ymm6","ymm7","ymm8","ymm9","ymm10","ymm11","ymm12","ymm13","ymm14","ymm15");
}
#undef D4SRC_FMA
__asm__(".space 128,0x90");
template<int S> static __attribute__((noinline)) void kpA(int n,const double *a,const double *b,double *c,int ldc,int rows, const double *L0,const double *L1,const double *L2,double *D0,double *D1,double *SD0) {
double *out=c;
ptrdiff_t stride=(ptrdiff_t)ldc*8;
int k=n;
ptrdiff_t idx=0;
asm volatile (
"vxorpd %%ymm0, %%ymm0, %%ymm0\n\t"
"vxorpd %%ymm1, %%ymm1, %%ymm1\n\t"
"vxorpd %%ymm2, %%ymm2, %%ymm2\n\t"
"vxorpd %%ymm3, %%ymm3, %%ymm3\n\t"
"vxorpd %%ymm4, %%ymm4, %%ymm4\n\t"
"vxorpd %%ymm5, %%ymm5, %%ymm5\n\t"
"vxorpd %%ymm6, %%ymm6, %%ymm6\n\t"
"vxorpd %%ymm7, %%ymm7, %%ymm7\n\t"
"vxorpd %%ymm8, %%ymm8, %%ymm8\n\t"
"vxorpd %%ymm9, %%ymm9, %%ymm9\n\t"
"vxorpd %%ymm10, %%ymm10, %%ymm10\n\t"
"vxorpd %%ymm11, %%ymm11, %%ymm11\n\t"
"vmovapd 0(%[b]), %%ymm12\n\t"
".p2align 6\n\t"
"1:\n\t"
"vmovapd 64(%[b]), %%ymm15\n\t"
"vmovapd 32(%[b]), %%ymm13\n\t"
"vbroadcastsd %c[o0]+0(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm0\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm1\n\t"
"vbroadcastsd %c[o1]+0(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm2\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm3\n\t"
"vbroadcastsd %c[o2]+0(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm4\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm5\n\t"
"vbroadcastsd %c[o3]+0(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm6\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm7\n\t"
"vbroadcastsd %c[o4]+0(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm8\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm9\n\t"
"vbroadcastsd %c[o5]+0(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm10\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm11\n\t"
"vmovapd 128(%[b]), %%ymm12\n\t"
"vmovapd 96(%[b]), %%ymm13\n\t"
"vbroadcastsd %c[o0]+8(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm0\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm1\n\t"
"vbroadcastsd %c[o1]+8(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm2\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm3\n\t"
"vbroadcastsd %c[o2]+8(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm4\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm5\n\t"
"vbroadcastsd %c[o3]+8(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm6\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm7\n\t"
"vbroadcastsd %c[o4]+8(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm8\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm9\n\t"
"vbroadcastsd %c[o5]+8(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm10\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm11\n\t"
"prefetcht0 512(%[b])\n\t"
"prefetcht0 576(%[b])\n\t"
"prefetcht0 640(%[b])\n\t"
"prefetcht0 704(%[b])\n\t"
"vmovapd 192(%[b]), %%ymm15\n\t"
"vmovapd 160(%[b]), %%ymm13\n\t"
"vbroadcastsd %c[o0]+16(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm0\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm1\n\t"
"vbroadcastsd %c[o1]+16(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm2\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm3\n\t"
"vbroadcastsd %c[o2]+16(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm4\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm5\n\t"
"vbroadcastsd %c[o3]+16(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm6\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm7\n\t"
"vbroadcastsd %c[o4]+16(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm8\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm9\n\t"
"vbroadcastsd %c[o5]+16(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm10\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm11\n\t"
"vmovapd 256(%[b]), %%ymm12\n\t"
"vmovapd 224(%[b]), %%ymm13\n\t"
"vbroadcastsd %c[o0]+24(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm0\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm1\n\t"
"vbroadcastsd %c[o1]+24(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm2\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm3\n\t"
"vbroadcastsd %c[o2]+24(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm4\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm5\n\t"
"vbroadcastsd %c[o3]+24(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm6\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm7\n\t"
"vbroadcastsd %c[o4]+24(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm8\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm9\n\t"
"vbroadcastsd %c[o5]+24(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm10\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm11\n\t"
"prefetcht0 32(%[L0],%[idx])\n\t"
"prefetcht0 64(%[L1],%[idx])\n\t"
"prefetcht0 96(%[L2],%[idx])\n\t"
"vmovupd (%[L0],%[idx]), %%ymm14\n\t"
"vmovupd (%[L2],%[idx]), %%ymm13\n\t"
"vmovupd (%[L1],%[idx]), %%ymm15\n\t"
"vsubpd %%ymm14, %%ymm13, %%ymm13\n\t"
"vmovntpd %%ymm13, (%[D1],%[idx])\n\t"
".if %c[s] == 1\n\t"
"vaddpd %%ymm15, %%ymm14, %%ymm14\n\t"
".else\n\t"
"vsubpd %%ymm15, %%ymm14, %%ymm14\n\t"
".endif\n\t"
"vmovntpd %%ymm14, (%[D0],%[idx])\n\t"
"vmovupd %%ymm14, (%[SD0],%[idx])\n\t""prefetcht0 768(%[b])\n\t"
"prefetcht0 832(%[b])\n\t"
"prefetcht0 896(%[b])\n\t"
"prefetcht0 960(%[b])\n\t"
"vmovapd 320(%[b]), %%ymm15\n\t"
"vmovapd 288(%[b]), %%ymm13\n\t"
"vbroadcastsd %c[o0]+32(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm0\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm1\n\t"
"vbroadcastsd %c[o1]+32(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm2\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm3\n\t"
"vbroadcastsd %c[o2]+32(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm4\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm5\n\t"
"vbroadcastsd %c[o3]+32(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm6\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm7\n\t"
"vbroadcastsd %c[o4]+32(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm8\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm9\n\t"
"vbroadcastsd %c[o5]+32(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm10\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm11\n\t"
"vmovapd 384(%[b]), %%ymm12\n\t"
"vmovapd 352(%[b]), %%ymm13\n\t"
"vbroadcastsd %c[o0]+40(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm0\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm1\n\t"
"vbroadcastsd %c[o1]+40(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm2\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm3\n\t"
"vbroadcastsd %c[o2]+40(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm4\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm5\n\t"
"vbroadcastsd %c[o3]+40(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm6\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm7\n\t"
"vbroadcastsd %c[o4]+40(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm8\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm9\n\t"
"vbroadcastsd %c[o5]+40(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm10\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm11\n\t"
"vmovapd 448(%[b]), %%ymm15\n\t"
"vmovapd 416(%[b]), %%ymm13\n\t"
"vbroadcastsd %c[o0]+48(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm0\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm1\n\t"
"vbroadcastsd %c[o1]+48(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm2\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm3\n\t"
"vbroadcastsd %c[o2]+48(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm4\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm5\n\t"
"vbroadcastsd %c[o3]+48(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm6\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm7\n\t"
"vbroadcastsd %c[o4]+48(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm8\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm9\n\t"
"vbroadcastsd %c[o5]+48(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm10\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm11\n\t"
"vmovapd 512(%[b]), %%ymm12\n\t"
"vmovapd 480(%[b]), %%ymm13\n\t"
"vbroadcastsd %c[o0]+56(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm0\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm1\n\t"
"vbroadcastsd %c[o1]+56(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm2\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm3\n\t"
"vbroadcastsd %c[o2]+56(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm4\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm5\n\t"
"vbroadcastsd %c[o3]+56(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm6\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm7\n\t"
"vbroadcastsd %c[o4]+56(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm8\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm9\n\t"
"vbroadcastsd %c[o5]+56(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm10\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm11\n\t"

"add $32, %[idx]\n\t"
"add $64, %[a]\n\t"
"add $512, %[b]\n\t"
"sub $8, %[k]\n\t"
"jnz 1b\n\t"
"vmovntpd %%ymm0, (%[out])\n\t"
"vmovntpd %%ymm1, 32(%[out])\n\t"
"vmovntpd %%ymm2, (%[out],%[stride])\n\t"
"vmovntpd %%ymm3, 32(%[out],%[stride])\n\t"
"vmovntpd %%ymm4, (%[out],%[stride],2)\n\t"
"vmovntpd %%ymm5, 32(%[out],%[stride],2)\n\t"
"lea (%[out],%[stride],2), %[out]\n\t"
"add %[stride], %[out]\n\t"
"vmovntpd %%ymm6, (%[out])\n\t"
"vmovntpd %%ymm7, 32(%[out])\n\t"
"vmovntpd %%ymm8, (%[out],%[stride])\n\t"
"vmovntpd %%ymm9, 32(%[out],%[stride])\n\t"
"vmovntpd %%ymm10, (%[out],%[stride],2)\n\t"
"vmovntpd %%ymm11, 32(%[out],%[stride],2)\n\t"
: [a] "+&r"(a), [b] "+&r"(b), [out] "+&r"(out), [k] "+&r"(k), [idx] "+&r"(idx)
: [L0] "r"(L0), [L1] "r"(L1), [L2] "r"(L2), [D0] "r"(D0), [D1] "r"(D1), [SD0] "r"(SD0), [s] "i"(S), [stride] "r"(stride), [o0] "i"(0*(BASE+24)*8), [o1] "i"(1*(BASE+24)*8), [o2] "i"(2*(BASE+24)*8), [o3] "i"(3*(BASE+24)*8), [o4] "i"(4*(BASE+24)*8), [o5] "i"(5*(BASE+24)*8)
: "memory", "cc", "ymm0","ymm1","ymm2","ymm3","ymm4","ymm5","ymm6","ymm7","ymm8","ymm9","ymm10","ymm11","ymm12","ymm13","ymm14","ymm15");
}
template<int S> static __attribute__((noinline)) void kpB(int n,const double *a,const double *b,double *c,int ldc,int rows, const double *L0,const double *L1,const double *L2,double *D0,double *D1) {
double *out=c;
ptrdiff_t stride=(ptrdiff_t)ldc*8;
int k=n;
ptrdiff_t idx=0;
asm volatile (
"vxorpd %%ymm0, %%ymm0, %%ymm0\n\t"
"vxorpd %%ymm1, %%ymm1, %%ymm1\n\t"
"vxorpd %%ymm2, %%ymm2, %%ymm2\n\t"
"vxorpd %%ymm3, %%ymm3, %%ymm3\n\t"
"vxorpd %%ymm4, %%ymm4, %%ymm4\n\t"
"vxorpd %%ymm5, %%ymm5, %%ymm5\n\t"
"vxorpd %%ymm6, %%ymm6, %%ymm6\n\t"
"vxorpd %%ymm7, %%ymm7, %%ymm7\n\t"
"vxorpd %%ymm8, %%ymm8, %%ymm8\n\t"
"vxorpd %%ymm9, %%ymm9, %%ymm9\n\t"
"vxorpd %%ymm10, %%ymm10, %%ymm10\n\t"
"vxorpd %%ymm11, %%ymm11, %%ymm11\n\t"
"vmovapd 0(%[b]), %%ymm12\n\t"
".p2align 6\n\t"
"1:\n\t"
"vmovapd 64(%[b]), %%ymm15\n\t"
"vmovapd 32(%[b]), %%ymm13\n\t"
"vbroadcastsd %c[o0]+0(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm0\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm1\n\t"
"vbroadcastsd %c[o1]+0(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm2\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm3\n\t"
"vbroadcastsd %c[o2]+0(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm4\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm5\n\t"
"vbroadcastsd %c[o3]+0(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm6\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm7\n\t"
"vbroadcastsd %c[o4]+0(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm8\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm9\n\t"
"vbroadcastsd %c[o5]+0(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm10\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm11\n\t"
"vmovapd 128(%[b]), %%ymm12\n\t"
"vmovapd 96(%[b]), %%ymm13\n\t"
"vbroadcastsd %c[o0]+8(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm0\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm1\n\t"
"vbroadcastsd %c[o1]+8(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm2\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm3\n\t"
"vbroadcastsd %c[o2]+8(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm4\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm5\n\t"
"vbroadcastsd %c[o3]+8(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm6\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm7\n\t"
"vbroadcastsd %c[o4]+8(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm8\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm9\n\t"
"vbroadcastsd %c[o5]+8(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm10\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm11\n\t"
"prefetcht0 512(%[b])\n\t"
"prefetcht0 576(%[b])\n\t"
"prefetcht0 640(%[b])\n\t"
"prefetcht0 704(%[b])\n\t"
"vmovapd 192(%[b]), %%ymm15\n\t"
"vmovapd 160(%[b]), %%ymm13\n\t"
"vbroadcastsd %c[o0]+16(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm0\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm1\n\t"
"vbroadcastsd %c[o1]+16(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm2\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm3\n\t"
"vbroadcastsd %c[o2]+16(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm4\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm5\n\t"
"vbroadcastsd %c[o3]+16(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm6\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm7\n\t"
"vbroadcastsd %c[o4]+16(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm8\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm9\n\t"
"vbroadcastsd %c[o5]+16(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm10\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm11\n\t"
"vmovapd 256(%[b]), %%ymm12\n\t"
"vmovapd 224(%[b]), %%ymm13\n\t"
"vbroadcastsd %c[o0]+24(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm0\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm1\n\t"
"vbroadcastsd %c[o1]+24(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm2\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm3\n\t"
"vbroadcastsd %c[o2]+24(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm4\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm5\n\t"
"vbroadcastsd %c[o3]+24(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm6\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm7\n\t"
"vbroadcastsd %c[o4]+24(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm8\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm9\n\t"
"vbroadcastsd %c[o5]+24(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm10\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm11\n\t"
"prefetcht0 32(%[L0],%[idx])\n\t"
"prefetcht0 64(%[L1],%[idx])\n\t"
"prefetcht0 96(%[L2],%[idx])\n\t"
"vmovupd (%[L0],%[idx]), %%ymm14\n\t"
"vmovupd (%[L1],%[idx]), %%ymm15\n\t"
"vmovupd (%[L2],%[idx]), %%ymm13\n\t"
"vsubpd %%ymm15, %%ymm14, %%ymm14\n\t"
"vmovntpd %%ymm14, (%[D0],%[idx])\n\t"
".if %c[s] == 0\n\t"
"vsubpd %%ymm14, %%ymm13, %%ymm13\n\t"
".else\n\t"
"vsubpd %%ymm13, %%ymm14, %%ymm13\n\t"
".endif\n\t"
"vmovntpd %%ymm13, (%[D1],%[idx])\n\t""prefetcht0 768(%[b])\n\t"
"prefetcht0 832(%[b])\n\t"
"prefetcht0 896(%[b])\n\t"
"prefetcht0 960(%[b])\n\t"
"vmovapd 320(%[b]), %%ymm15\n\t"
"vmovapd 288(%[b]), %%ymm13\n\t"
"vbroadcastsd %c[o0]+32(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm0\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm1\n\t"
"vbroadcastsd %c[o1]+32(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm2\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm3\n\t"
"vbroadcastsd %c[o2]+32(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm4\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm5\n\t"
"vbroadcastsd %c[o3]+32(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm6\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm7\n\t"
"vbroadcastsd %c[o4]+32(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm8\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm9\n\t"
"vbroadcastsd %c[o5]+32(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm10\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm11\n\t"
"vmovapd 384(%[b]), %%ymm12\n\t"
"vmovapd 352(%[b]), %%ymm13\n\t"
"vbroadcastsd %c[o0]+40(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm0\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm1\n\t"
"vbroadcastsd %c[o1]+40(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm2\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm3\n\t"
"vbroadcastsd %c[o2]+40(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm4\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm5\n\t"
"vbroadcastsd %c[o3]+40(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm6\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm7\n\t"
"vbroadcastsd %c[o4]+40(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm8\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm9\n\t"
"vbroadcastsd %c[o5]+40(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm10\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm11\n\t"
"vmovapd 448(%[b]), %%ymm15\n\t"
"vmovapd 416(%[b]), %%ymm13\n\t"
"vbroadcastsd %c[o0]+48(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm0\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm1\n\t"
"vbroadcastsd %c[o1]+48(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm2\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm3\n\t"
"vbroadcastsd %c[o2]+48(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm4\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm5\n\t"
"vbroadcastsd %c[o3]+48(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm6\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm7\n\t"
"vbroadcastsd %c[o4]+48(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm8\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm9\n\t"
"vbroadcastsd %c[o5]+48(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm10\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm11\n\t"
"vmovapd 512(%[b]), %%ymm12\n\t"
"vmovapd 480(%[b]), %%ymm13\n\t"
"vbroadcastsd %c[o0]+56(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm0\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm1\n\t"
"vbroadcastsd %c[o1]+56(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm2\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm3\n\t"
"vbroadcastsd %c[o2]+56(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm4\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm5\n\t"
"vbroadcastsd %c[o3]+56(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm6\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm7\n\t"
"vbroadcastsd %c[o4]+56(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm8\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm9\n\t"
"vbroadcastsd %c[o5]+56(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm10\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm11\n\t"

"add $32, %[idx]\n\t"
"add $64, %[a]\n\t"
"add $512, %[b]\n\t"
"sub $8, %[k]\n\t"
"jnz 1b\n\t"
"vmovntpd %%ymm0, (%[out])\n\t"
"vmovntpd %%ymm1, 32(%[out])\n\t"
"vmovntpd %%ymm2, (%[out],%[stride])\n\t"
"vmovntpd %%ymm3, 32(%[out],%[stride])\n\t"
"vmovntpd %%ymm4, (%[out],%[stride],2)\n\t"
"vmovntpd %%ymm5, 32(%[out],%[stride],2)\n\t"
"lea (%[out],%[stride],2), %[out]\n\t"
"add %[stride], %[out]\n\t"
"vmovntpd %%ymm6, (%[out])\n\t"
"vmovntpd %%ymm7, 32(%[out])\n\t"
"vmovntpd %%ymm8, (%[out],%[stride])\n\t"
"vmovntpd %%ymm9, 32(%[out],%[stride])\n\t"
"vmovntpd %%ymm10, (%[out],%[stride],2)\n\t"
"vmovntpd %%ymm11, 32(%[out],%[stride],2)\n\t"
: [a] "+&r"(a), [b] "+&r"(b), [out] "+&r"(out), [k] "+&r"(k), [idx] "+&r"(idx)
: [L0] "r"(L0), [L1] "r"(L1), [L2] "r"(L2), [D0] "r"(D0), [D1] "r"(D1), [s] "i"(S), [stride] "r"(stride), [o0] "i"(0*(BASE+24)*8), [o1] "i"(1*(BASE+24)*8), [o2] "i"(2*(BASE+24)*8), [o3] "i"(3*(BASE+24)*8), [o4] "i"(4*(BASE+24)*8), [o5] "i"(5*(BASE+24)*8)
: "memory", "cc", "ymm0","ymm1","ymm2","ymm3","ymm4","ymm5","ymm6","ymm7","ymm8","ymm9","ymm10","ymm11","ymm12","ymm13","ymm14","ymm15");
}
static __attribute__((always_inline)) inline void kernel_plain(int n,const double *a,const double *b,double *c,int ldc,int rows) {
double *out=c;
ptrdiff_t stride=(ptrdiff_t)ldc*8;
int k=n;
asm volatile (
"vxorpd %%ymm0, %%ymm0, %%ymm0\n\t"
"vxorpd %%ymm1, %%ymm1, %%ymm1\n\t"
"vxorpd %%ymm2, %%ymm2, %%ymm2\n\t"
"vxorpd %%ymm3, %%ymm3, %%ymm3\n\t"
"vxorpd %%ymm4, %%ymm4, %%ymm4\n\t"
"vxorpd %%ymm5, %%ymm5, %%ymm5\n\t"
"vxorpd %%ymm6, %%ymm6, %%ymm6\n\t"
"vxorpd %%ymm7, %%ymm7, %%ymm7\n\t"
"vxorpd %%ymm8, %%ymm8, %%ymm8\n\t"
"vxorpd %%ymm9, %%ymm9, %%ymm9\n\t"
"vxorpd %%ymm10, %%ymm10, %%ymm10\n\t"
"vxorpd %%ymm11, %%ymm11, %%ymm11\n\t"
"vmovapd 0(%[b]), %%ymm12\n\t"
".p2align 6\n\t"
"1:\n\t"
"vmovapd 64(%[b]), %%ymm15\n\t"
"vmovapd 32(%[b]), %%ymm13\n\t"
"vbroadcastsd %c[o0]+0(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm0\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm1\n\t"
"vbroadcastsd %c[o1]+0(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm2\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm3\n\t"
"vbroadcastsd %c[o2]+0(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm4\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm5\n\t"
"vbroadcastsd %c[o3]+0(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm6\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm7\n\t"
"vbroadcastsd %c[o4]+0(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm8\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm9\n\t"
"vbroadcastsd %c[o5]+0(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm10\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm11\n\t"
"vmovapd 128(%[b]), %%ymm12\n\t"
"vmovapd 96(%[b]), %%ymm13\n\t"
"vbroadcastsd %c[o0]+8(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm0\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm1\n\t"
"vbroadcastsd %c[o1]+8(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm2\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm3\n\t"
"vbroadcastsd %c[o2]+8(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm4\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm5\n\t"
"vbroadcastsd %c[o3]+8(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm6\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm7\n\t"
"vbroadcastsd %c[o4]+8(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm8\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm9\n\t"
"vbroadcastsd %c[o5]+8(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm10\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm11\n\t"
"prefetcht0 512(%[b])\n\t"
"prefetcht0 576(%[b])\n\t"
"prefetcht0 640(%[b])\n\t"
"prefetcht0 704(%[b])\n\t"
"vmovapd 192(%[b]), %%ymm15\n\t"
"vmovapd 160(%[b]), %%ymm13\n\t"
"vbroadcastsd %c[o0]+16(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm0\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm1\n\t"
"vbroadcastsd %c[o1]+16(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm2\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm3\n\t"
"vbroadcastsd %c[o2]+16(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm4\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm5\n\t"
"vbroadcastsd %c[o3]+16(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm6\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm7\n\t"
"vbroadcastsd %c[o4]+16(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm8\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm9\n\t"
"vbroadcastsd %c[o5]+16(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm10\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm11\n\t"
"vmovapd 256(%[b]), %%ymm12\n\t"
"vmovapd 224(%[b]), %%ymm13\n\t"
"vbroadcastsd %c[o0]+24(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm0\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm1\n\t"
"vbroadcastsd %c[o1]+24(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm2\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm3\n\t"
"vbroadcastsd %c[o2]+24(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm4\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm5\n\t"
"vbroadcastsd %c[o3]+24(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm6\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm7\n\t"
"vbroadcastsd %c[o4]+24(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm8\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm9\n\t"
"vbroadcastsd %c[o5]+24(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm10\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm11\n\t"
"prefetcht0 768(%[b])\n\t"
"prefetcht0 832(%[b])\n\t"
"prefetcht0 896(%[b])\n\t"
"prefetcht0 960(%[b])\n\t"
"vmovapd 320(%[b]), %%ymm15\n\t"
"vmovapd 288(%[b]), %%ymm13\n\t"
"vbroadcastsd %c[o0]+32(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm0\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm1\n\t"
"vbroadcastsd %c[o1]+32(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm2\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm3\n\t"
"vbroadcastsd %c[o2]+32(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm4\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm5\n\t"
"vbroadcastsd %c[o3]+32(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm6\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm7\n\t"
"vbroadcastsd %c[o4]+32(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm8\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm9\n\t"
"vbroadcastsd %c[o5]+32(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm10\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm11\n\t"
"vmovapd 384(%[b]), %%ymm12\n\t"
"vmovapd 352(%[b]), %%ymm13\n\t"
"vbroadcastsd %c[o0]+40(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm0\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm1\n\t"
"vbroadcastsd %c[o1]+40(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm2\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm3\n\t"
"vbroadcastsd %c[o2]+40(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm4\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm5\n\t"
"vbroadcastsd %c[o3]+40(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm6\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm7\n\t"
"vbroadcastsd %c[o4]+40(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm8\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm9\n\t"
"vbroadcastsd %c[o5]+40(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm10\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm11\n\t"
"vmovapd 448(%[b]), %%ymm15\n\t"
"vmovapd 416(%[b]), %%ymm13\n\t"
"vbroadcastsd %c[o0]+48(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm0\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm1\n\t"
"vbroadcastsd %c[o1]+48(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm2\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm3\n\t"
"vbroadcastsd %c[o2]+48(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm4\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm5\n\t"
"vbroadcastsd %c[o3]+48(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm6\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm7\n\t"
"vbroadcastsd %c[o4]+48(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm8\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm9\n\t"
"vbroadcastsd %c[o5]+48(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm10\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm11\n\t"
"vmovapd 512(%[b]), %%ymm12\n\t"
"vmovapd 480(%[b]), %%ymm13\n\t"
"vbroadcastsd %c[o0]+56(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm0\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm1\n\t"
"vbroadcastsd %c[o1]+56(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm2\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm3\n\t"
"vbroadcastsd %c[o2]+56(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm4\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm5\n\t"
"vbroadcastsd %c[o3]+56(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm6\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm7\n\t"
"vbroadcastsd %c[o4]+56(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm8\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm9\n\t"
"vbroadcastsd %c[o5]+56(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm10\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm11\n\t"
"add $64, %[a]\n\t"
"add $512, %[b]\n\t"
"sub $8, %[k]\n\t"
"jnz 1b\n\t"
"vmovntpd %%ymm0, (%[out])\n\t"
"vmovntpd %%ymm1, 32(%[out])\n\t"
"vmovntpd %%ymm2, (%[out],%[stride])\n\t"
"vmovntpd %%ymm3, 32(%[out],%[stride])\n\t"
"vmovntpd %%ymm4, (%[out],%[stride],2)\n\t"
"vmovntpd %%ymm5, 32(%[out],%[stride],2)\n\t"
"lea (%[out],%[stride],2), %[out]\n\t"
"add %[stride], %[out]\n\t"
"vmovntpd %%ymm6, (%[out])\n\t"
"vmovntpd %%ymm7, 32(%[out])\n\t"
"vmovntpd %%ymm8, (%[out],%[stride])\n\t"
"vmovntpd %%ymm9, 32(%[out],%[stride])\n\t"
"vmovntpd %%ymm10, (%[out],%[stride],2)\n\t"
"vmovntpd %%ymm11, 32(%[out],%[stride],2)\n\t"
: [a] "+&r"(a), [b] "+&r"(b), [out] "+&r"(out), [k] "+&r"(k)
: [stride] "r"(stride), [o0] "i"(0*(BASE+24)*8), [o1] "i"(1*(BASE+24)*8), [o2] "i"(2*(BASE+24)*8), [o3] "i"(3*(BASE+24)*8), [o4] "i"(4*(BASE+24)*8), [o5] "i"(5*(BASE+24)*8)
: "memory", "cc", "ymm0","ymm1","ymm2","ymm3","ymm4","ymm5","ymm6","ymm7","ymm8","ymm9","ymm10","ymm11","ymm12","ymm13","ymm14","ymm15");
}
static __attribute__((always_inline)) inline void kernel_reverse(int n,const double *a,const double *b,double *c,int ldc,int rows) {
a += n-1; b += (size_t)(n-1)*8;
double *out=c;
ptrdiff_t stride=(ptrdiff_t)ldc*8;
int k=n;
asm volatile (
"vxorpd %%ymm0, %%ymm0, %%ymm0\n\t"
"vxorpd %%ymm1, %%ymm1, %%ymm1\n\t"
"vxorpd %%ymm2, %%ymm2, %%ymm2\n\t"
"vxorpd %%ymm3, %%ymm3, %%ymm3\n\t"
"vxorpd %%ymm4, %%ymm4, %%ymm4\n\t"
"vxorpd %%ymm5, %%ymm5, %%ymm5\n\t"
"vxorpd %%ymm6, %%ymm6, %%ymm6\n\t"
"vxorpd %%ymm7, %%ymm7, %%ymm7\n\t"
"vxorpd %%ymm8, %%ymm8, %%ymm8\n\t"
"vxorpd %%ymm9, %%ymm9, %%ymm9\n\t"
"vxorpd %%ymm10, %%ymm10, %%ymm10\n\t"
"vxorpd %%ymm11, %%ymm11, %%ymm11\n\t"
"vmovapd 0(%[b]), %%ymm12\n\t"
".p2align 6\n\t"
"1:\n\t"
"vmovapd -64(%[b]), %%ymm15\n\t"
"vmovapd 32(%[b]), %%ymm13\n\t"
"vbroadcastsd %c[o0]-0(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm0\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm1\n\t"
"vbroadcastsd %c[o1]-0(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm2\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm3\n\t"
"vbroadcastsd %c[o2]-0(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm4\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm5\n\t"
"vbroadcastsd %c[o3]-0(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm6\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm7\n\t"
"vbroadcastsd %c[o4]-0(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm8\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm9\n\t"
"vbroadcastsd %c[o5]-0(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm10\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm11\n\t"
"vmovapd -128(%[b]), %%ymm12\n\t"
"vmovapd -32(%[b]), %%ymm13\n\t"
"vbroadcastsd %c[o0]-8(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm0\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm1\n\t"
"vbroadcastsd %c[o1]-8(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm2\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm3\n\t"
"vbroadcastsd %c[o2]-8(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm4\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm5\n\t"
"vbroadcastsd %c[o3]-8(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm6\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm7\n\t"
"vbroadcastsd %c[o4]-8(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm8\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm9\n\t"
"vbroadcastsd %c[o5]-8(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm10\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm11\n\t"
"prefetcht0 -512(%[b])\n\t"
"prefetcht0 -576(%[b])\n\t"
"prefetcht0 -640(%[b])\n\t"
"prefetcht0 -704(%[b])\n\t"
"vmovapd -192(%[b]), %%ymm15\n\t"
"vmovapd -96(%[b]), %%ymm13\n\t"
"vbroadcastsd %c[o0]-16(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm0\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm1\n\t"
"vbroadcastsd %c[o1]-16(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm2\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm3\n\t"
"vbroadcastsd %c[o2]-16(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm4\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm5\n\t"
"vbroadcastsd %c[o3]-16(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm6\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm7\n\t"
"vbroadcastsd %c[o4]-16(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm8\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm9\n\t"
"vbroadcastsd %c[o5]-16(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm10\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm11\n\t"
"vmovapd -256(%[b]), %%ymm12\n\t"
"vmovapd -160(%[b]), %%ymm13\n\t"
"vbroadcastsd %c[o0]-24(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm0\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm1\n\t"
"vbroadcastsd %c[o1]-24(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm2\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm3\n\t"
"vbroadcastsd %c[o2]-24(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm4\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm5\n\t"
"vbroadcastsd %c[o3]-24(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm6\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm7\n\t"
"vbroadcastsd %c[o4]-24(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm8\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm9\n\t"
"vbroadcastsd %c[o5]-24(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm10\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm11\n\t"
"prefetcht0 -768(%[b])\n\t"
"prefetcht0 -832(%[b])\n\t"
"prefetcht0 -896(%[b])\n\t"
"prefetcht0 -960(%[b])\n\t"
"vmovapd -320(%[b]), %%ymm15\n\t"
"vmovapd -224(%[b]), %%ymm13\n\t"
"vbroadcastsd %c[o0]-32(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm0\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm1\n\t"
"vbroadcastsd %c[o1]-32(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm2\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm3\n\t"
"vbroadcastsd %c[o2]-32(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm4\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm5\n\t"
"vbroadcastsd %c[o3]-32(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm6\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm7\n\t"
"vbroadcastsd %c[o4]-32(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm8\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm9\n\t"
"vbroadcastsd %c[o5]-32(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm10\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm11\n\t"
"vmovapd -384(%[b]), %%ymm12\n\t"
"vmovapd -288(%[b]), %%ymm13\n\t"
"vbroadcastsd %c[o0]-40(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm0\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm1\n\t"
"vbroadcastsd %c[o1]-40(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm2\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm3\n\t"
"vbroadcastsd %c[o2]-40(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm4\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm5\n\t"
"vbroadcastsd %c[o3]-40(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm6\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm7\n\t"
"vbroadcastsd %c[o4]-40(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm8\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm9\n\t"
"vbroadcastsd %c[o5]-40(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm10\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm11\n\t"
"vmovapd -448(%[b]), %%ymm15\n\t"
"vmovapd -352(%[b]), %%ymm13\n\t"
"vbroadcastsd %c[o0]-48(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm0\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm1\n\t"
"vbroadcastsd %c[o1]-48(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm2\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm3\n\t"
"vbroadcastsd %c[o2]-48(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm4\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm5\n\t"
"vbroadcastsd %c[o3]-48(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm6\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm7\n\t"
"vbroadcastsd %c[o4]-48(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm8\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm9\n\t"
"vbroadcastsd %c[o5]-48(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm10\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm11\n\t"
"vmovapd -512(%[b]), %%ymm12\n\t"
"vmovapd -416(%[b]), %%ymm13\n\t"
"vbroadcastsd %c[o0]-56(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm0\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm1\n\t"
"vbroadcastsd %c[o1]-56(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm2\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm3\n\t"
"vbroadcastsd %c[o2]-56(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm4\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm5\n\t"
"vbroadcastsd %c[o3]-56(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm6\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm7\n\t"
"vbroadcastsd %c[o4]-56(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm8\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm9\n\t"
"vbroadcastsd %c[o5]-56(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm15, %%ymm14, %%ymm10\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm11\n\t"
"sub $64, %[a]\n\t"
"sub $512, %[b]\n\t"
"sub $8, %[k]\n\t"
"jnz 1b\n\t"
"vmovntpd %%ymm0, (%[out])\n\t"
"vmovntpd %%ymm1, 32(%[out])\n\t"
"vmovntpd %%ymm2, (%[out],%[stride])\n\t"
"vmovntpd %%ymm3, 32(%[out],%[stride])\n\t"
"vmovntpd %%ymm4, (%[out],%[stride],2)\n\t"
"vmovntpd %%ymm5, 32(%[out],%[stride],2)\n\t"
"lea (%[out],%[stride],2), %[out]\n\t"
"add %[stride], %[out]\n\t"
"vmovntpd %%ymm6, (%[out])\n\t"
"vmovntpd %%ymm7, 32(%[out])\n\t"
"vmovntpd %%ymm8, (%[out],%[stride])\n\t"
"vmovntpd %%ymm9, 32(%[out],%[stride])\n\t"
"vmovntpd %%ymm10, (%[out],%[stride],2)\n\t"
"vmovntpd %%ymm11, 32(%[out],%[stride],2)\n\t"
: [a] "+&r"(a), [b] "+&r"(b), [out] "+&r"(out), [k] "+&r"(k)
: [stride] "r"(stride), [o0] "i"(0*(BASE+24)*8), [o1] "i"(1*(BASE+24)*8), [o2] "i"(2*(BASE+24)*8), [o3] "i"(3*(BASE+24)*8), [o4] "i"(4*(BASE+24)*8), [o5] "i"(5*(BASE+24)*8)
: "memory", "cc", "ymm0","ymm1","ymm2","ymm3","ymm4","ymm5","ymm6","ymm7","ymm8","ymm9","ymm10","ymm11","ymm12","ymm13","ymm14","ymm15");
}
struct Pend2 { double *c; int ldc; const double *m; int h; int pz; int pos; bool active; const double *add=nullptr; bool part=false; };
static Pend2 pend2={0,0,0,0,0,0,false};
static constexpr int STEP2=16;
template<int H=1024> static inline void pend2_prefetch(void){
if(!pend2.active)return;
if(pend2.part){                      
int i=pend2.pos>>10, j=pend2.pos&1023;
const double *c=pend2.c+(size_t)i*pend2.ldc+j;
const double *mm=pend2.m+pend2.pos;
const double *pp=pend2.add+(size_t)i*2048+j;
for(int t=0;t<64;t+=8) {
_mm_prefetch((const char*)(c+t),_MM_HINT_T1);
_mm_prefetch((const char*)(c+1024+t),_MM_HINT_T1);
_mm_prefetch((const char*)(c+(size_t)1024*pend2.ldc+t),_MM_HINT_T1);
_mm_prefetch((const char*)(c+(size_t)1024*pend2.ldc+1024+t),_MM_HINT_T1);
_mm_prefetch((const char*)(mm+t),_MM_HINT_T1);
_mm_prefetch((const char*)(mm+(1024*1024+UPPER_PAGE_PAD)+t),_MM_HINT_T1);
_mm_prefetch((const char*)(mm+2*(1024*1024+UPPER_PAGE_PAD)+t),_MM_HINT_T1);
_mm_prefetch((const char*)(pp+t),_MM_HINT_T1);
_mm_prefetch((const char*)(pp+(size_t)1024*2048+t),_MM_HINT_T1);
_mm_prefetch((const char*)(pp+(size_t)1024*2048+1024+t),_MM_HINT_T1);
}
return;
}
int i=pend2.pos >> __builtin_ctz((unsigned)H), j=pend2.pos & (H-1);
const double *c=pend2.c+(size_t)i*pend2.ldc+j;
const double *m=pend2.m+pend2.pos;
for(int t=0;t<STEP2;t+=8) {
_mm_prefetch((const char*)(c+t),_MM_HINT_T1);
_mm_prefetch((const char*)(c+H+t),_MM_HINT_T1);
_mm_prefetch((const char*)(c+(size_t)H*pend2.ldc+t),_MM_HINT_T1);
_mm_prefetch((const char*)(c+(size_t)H*pend2.ldc+H+t),_MM_HINT_T1);
_mm_prefetch((const char*)(m+t),_MM_HINT_T1);
_mm_prefetch((const char*)(m+(H*H+UPPER_PAGE_PAD)+t),_MM_HINT_T1);
_mm_prefetch((const char*)(m+2*(H*H+UPPER_PAGE_PAD)+t),_MM_HINT_T1);
}
}
template<int H=1024,bool ADD=false> static inline void pend2_step(void){
int i=pend2.pos >> __builtin_ctz((unsigned)H), j=pend2.pos & (H-1);
for(int t=0;t<STEP2;t+=4) {
double *c11=pend2.c+(size_t)i*pend2.ldc+j+t,*c12=c11+H,
*c21=c11+(size_t)H*pend2.ldc,*c22=c21+H;
const double *m=pend2.m+pend2.pos+t;
__m256d m1=_mm256_load_pd(c11),m2=_mm256_load_pd(c21),
m3=_mm256_load_pd(c12),m4=_mm256_load_pd(c22),
m5=_mm256_load_pd(m),m6=_mm256_load_pd(m+(H*H+UPPER_PAGE_PAD)),m7=_mm256_load_pd(m+2*(H*H+UPPER_PAGE_PAD));
__m256d u2=_mm256_add_pd(m1,m6),u3=_mm256_add_pd(u2,m7),u4=_mm256_add_pd(u2,m5);
__m256d r11=_mm256_add_pd(m1,m2),r12=_mm256_add_pd(u4,m3),
r21=_mm256_sub_pd(u3,m4),r22=_mm256_add_pd(u3,m5);
if(ADD) {
const double *p=pend2.add+(size_t)i*2048+j+t;
r11=_mm256_add_pd(_mm256_load_pd(p),r11);
r12=_mm256_add_pd(_mm256_load_pd(p+H),r12);
r21=_mm256_add_pd(_mm256_load_pd(p+(size_t)H*2048),r21);
r22=_mm256_add_pd(_mm256_load_pd(p+(size_t)H*2048+H),r22);
}
_mm256_store_pd(c11,r11);_mm256_store_pd(c12,r12);
_mm256_store_pd(c21,r21);_mm256_store_pd(c22,r22);
}
pend2.pos+=STEP2;
if(pend2.pos>=H*H)pend2.active=false;
}
static void pend2_finish(void){
if(pend2.add) {while(pend2.active)pend2_step<1024,true>();}
else if(pend2.h==1024) {while(pend2.active)pend2_step<1024>();}
else {while(pend2.active)pend2_step<2048>();}
}
struct Pending { double *c; int ldc; const double *m; int pos=0; bool active=false; int rowlim=0; bool prog=false; };
static Pending pending;
alignas(4096) static double pending_work[3*(BASE*BASE+LEAF_PAD)+LEAF_PAD];
static unsigned pending_tick=0;
static constexpr int STEP=64, PERIOD=1;
static inline void pending_prefetch() {
if(!pending.active)return;
if(pending.prog && pending.pos/BASE>=pending.rowlim)return;
int i=pending.pos/BASE,j=pending.pos%BASE;
const double *c=pending.c+(size_t)i*pending.ldc+j;
const double *m=pending.m+pending.pos;
for(int t=0;t<STEP;t+=8) {
_mm_prefetch((const char*)(c+t),_MM_HINT_T1);
_mm_prefetch((const char*)(c+BASE+t),_MM_HINT_T1);
_mm_prefetch((const char*)(c+(size_t)BASE*pending.ldc+t),_MM_HINT_T1);
_mm_prefetch((const char*)(c+(size_t)BASE*pending.ldc+BASE+t),_MM_HINT_T1);
_mm_prefetch((const char*)(m+t),_MM_HINT_T1);
_mm_prefetch((const char*)(m+(BASE*BASE+LEAF_PAD)+t),_MM_HINT_T1);
_mm_prefetch((const char*)(m+2*(BASE*BASE+LEAF_PAD)+t),_MM_HINT_T1);
}
}
static inline void pending_step_unchecked() {
int i=pending.pos/BASE,j=pending.pos%BASE;
for(int t=0;t<STEP;t+=4) {
double *c11=pending.c+(size_t)i*pending.ldc+j+t,
*c12=c11+BASE,*c21=c11+(size_t)BASE*pending.ldc,*c22=c21+BASE;
const double *m=pending.m+pending.pos+t;
__m256d m1=_mm256_load_pd(c11),m2=_mm256_load_pd(c21),
m3=_mm256_load_pd(c12),m4=_mm256_load_pd(c22),
m5=_mm256_load_pd(m),m6=_mm256_load_pd(m+(BASE*BASE+LEAF_PAD)),m7=_mm256_load_pd(m+2*(BASE*BASE+LEAF_PAD));
__m256d r11=_mm256_add_pd(_mm256_sub_pd(_mm256_add_pd(m1,m4),m5),m7),
r12=_mm256_add_pd(m3,m5),r21=_mm256_add_pd(m2,m4),
r22=_mm256_add_pd(_mm256_add_pd(_mm256_sub_pd(m1,m2),m3),m6);
_mm256_store_pd(c11,r11);_mm256_store_pd(c12,r12);
_mm256_store_pd(c21,r21);_mm256_store_pd(c22,r22);
}
pending.pos+=STEP;
if(pending.pos==BASE*BASE)pending.active=false;
}
static inline void pending_step(){if(pending.prog && pending.pos/BASE>=pending.rowlim)return;pending_step_unchecked();}
static void pending_finish() {
pending.prog=false;
while(pending.active)pending_step();
}
static __attribute__((always_inline)) inline void kernel2(int n,const double *a,const double *b,double *c,int ldc) {
int k=n;
double *out=c;
ptrdiff_t stride=(ptrdiff_t)ldc*8;
asm volatile (
"vxorpd %%ymm0, %%ymm0, %%ymm0\n\t"
"vxorpd %%ymm1, %%ymm1, %%ymm1\n\t"
"vxorpd %%ymm2, %%ymm2, %%ymm2\n\t"
"vxorpd %%ymm3, %%ymm3, %%ymm3\n\t"
".p2align 6\n\t" "1:\n\t"
"prefetcht0 512(%[b])\n\t"
"prefetcht0 576(%[b])\n\t"
"prefetcht0 640(%[b])\n\t"
"prefetcht0 704(%[b])\n\t"
"vmovapd 0(%[b]), %%ymm12\n\t"
"vmovapd 32(%[b]), %%ymm13\n\t"
"vbroadcastsd %c[o0]+0(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm0\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm1\n\t"
"vbroadcastsd %c[o1]+0(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm2\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm3\n\t"
"vmovapd 64(%[b]), %%ymm12\n\t"
"vmovapd 96(%[b]), %%ymm13\n\t"
"vbroadcastsd %c[o0]+8(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm0\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm1\n\t"
"vbroadcastsd %c[o1]+8(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm2\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm3\n\t"
"vmovapd 128(%[b]), %%ymm12\n\t"
"vmovapd 160(%[b]), %%ymm13\n\t"
"vbroadcastsd %c[o0]+16(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm0\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm1\n\t"
"vbroadcastsd %c[o1]+16(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm2\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm3\n\t"
"vmovapd 192(%[b]), %%ymm12\n\t"
"vmovapd 224(%[b]), %%ymm13\n\t"
"vbroadcastsd %c[o0]+24(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm0\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm1\n\t"
"vbroadcastsd %c[o1]+24(%[a]), %%ymm14\n\t"
"vfmadd231pd %%ymm12, %%ymm14, %%ymm2\n\t"
"vfmadd231pd %%ymm13, %%ymm14, %%ymm3\n\t"
"add $32, %[a]\n\t"
"add $256, %[b]\n\t"
"sub $4, %[k]\n\t"
"jnz 1b\n\t"
"vmovntpd %%ymm0, (%[out])\n\t"
"vmovntpd %%ymm1, 32(%[out])\n\t"
"add %[stride], %[out]\n\t"
"vmovntpd %%ymm2, (%[out])\n\t"
"vmovntpd %%ymm3, 32(%[out])\n\t"
: [a] "+&r"(a), [b] "+&r"(b), [out] "+&r"(out), [k] "+&r"(k)
: [stride] "r"(stride), [o0] "i"(0*(BASE+24)*8), [o1] "i"(1*(BASE+24)*8)
: "memory", "cc", "ymm0","ymm1","ymm2","ymm3","ymm12","ymm13","ymm14");
}
static __attribute__((noinline)) void kernel_tail24(int n,const double *a,const double *b,double *out,int ldc) {
const double *b1=b+8*(n+1),*b2=b+16*(n+1);
ptrdiff_t stride=(ptrdiff_t)ldc*8; int kk=n;
asm volatile (
"vxorpd %%ymm0, %%ymm0, %%ymm0\n\t"
"vxorpd %%ymm1, %%ymm1, %%ymm1\n\t"
"vxorpd %%ymm2, %%ymm2, %%ymm2\n\t"
"vxorpd %%ymm3, %%ymm3, %%ymm3\n\t"
"vxorpd %%ymm4, %%ymm4, %%ymm4\n\t"
"vxorpd %%ymm5, %%ymm5, %%ymm5\n\t"
"vxorpd %%ymm6, %%ymm6, %%ymm6\n\t"
"vxorpd %%ymm7, %%ymm7, %%ymm7\n\t"
"vxorpd %%ymm8, %%ymm8, %%ymm8\n\t"
"vxorpd %%ymm9, %%ymm9, %%ymm9\n\t"
"vxorpd %%ymm10, %%ymm10, %%ymm10\n\t"
"vxorpd %%ymm11, %%ymm11, %%ymm11\n\t"
".p2align 6\n\t"
"1:\n\t"
"prefetcht0 512(%[b])\n\t"
"prefetcht0 512(%[b1])\n\t"
"prefetcht0 512(%[b2])\n\t"
"vbroadcastsd %c[o0]+0(%[a]), %%ymm12\n\t"
"vbroadcastsd %c[o1]+0(%[a]), %%ymm13\n\t"
"vmovapd 0(%[b]), %%ymm14\n\t"
"vfmadd231pd %%ymm14, %%ymm12, %%ymm0\n\t"
"vfmadd231pd %%ymm14, %%ymm13, %%ymm6\n\t"
"vmovapd 32(%[b]), %%ymm14\n\t"
"vfmadd231pd %%ymm14, %%ymm12, %%ymm1\n\t"
"vfmadd231pd %%ymm14, %%ymm13, %%ymm7\n\t"
"vmovapd 0(%[b1]), %%ymm14\n\t"
"vfmadd231pd %%ymm14, %%ymm12, %%ymm2\n\t"
"vfmadd231pd %%ymm14, %%ymm13, %%ymm8\n\t"
"vmovapd 32(%[b1]), %%ymm14\n\t"
"vfmadd231pd %%ymm14, %%ymm12, %%ymm3\n\t"
"vfmadd231pd %%ymm14, %%ymm13, %%ymm9\n\t"
"vmovapd 0(%[b2]), %%ymm14\n\t"
"vfmadd231pd %%ymm14, %%ymm12, %%ymm4\n\t"
"vfmadd231pd %%ymm14, %%ymm13, %%ymm10\n\t"
"vmovapd 32(%[b2]), %%ymm14\n\t"
"vfmadd231pd %%ymm14, %%ymm12, %%ymm5\n\t"
"vfmadd231pd %%ymm14, %%ymm13, %%ymm11\n\t"
"prefetcht0 576(%[b])\n\t"
"prefetcht0 576(%[b1])\n\t"
"prefetcht0 576(%[b2])\n\t"
"vbroadcastsd %c[o0]+8(%[a]), %%ymm12\n\t"
"vbroadcastsd %c[o1]+8(%[a]), %%ymm13\n\t"
"vmovapd 64(%[b]), %%ymm14\n\t"
"vfmadd231pd %%ymm14, %%ymm12, %%ymm0\n\t"
"vfmadd231pd %%ymm14, %%ymm13, %%ymm6\n\t"
"vmovapd 96(%[b]), %%ymm14\n\t"
"vfmadd231pd %%ymm14, %%ymm12, %%ymm1\n\t"
"vfmadd231pd %%ymm14, %%ymm13, %%ymm7\n\t"
"vmovapd 64(%[b1]), %%ymm14\n\t"
"vfmadd231pd %%ymm14, %%ymm12, %%ymm2\n\t"
"vfmadd231pd %%ymm14, %%ymm13, %%ymm8\n\t"
"vmovapd 96(%[b1]), %%ymm14\n\t"
"vfmadd231pd %%ymm14, %%ymm12, %%ymm3\n\t"
"vfmadd231pd %%ymm14, %%ymm13, %%ymm9\n\t"
"vmovapd 64(%[b2]), %%ymm14\n\t"
"vfmadd231pd %%ymm14, %%ymm12, %%ymm4\n\t"
"vfmadd231pd %%ymm14, %%ymm13, %%ymm10\n\t"
"vmovapd 96(%[b2]), %%ymm14\n\t"
"vfmadd231pd %%ymm14, %%ymm12, %%ymm5\n\t"
"vfmadd231pd %%ymm14, %%ymm13, %%ymm11\n\t"
"prefetcht0 640(%[b])\n\t"
"prefetcht0 640(%[b1])\n\t"
"prefetcht0 640(%[b2])\n\t"
"vbroadcastsd %c[o0]+16(%[a]), %%ymm12\n\t"
"vbroadcastsd %c[o1]+16(%[a]), %%ymm13\n\t"
"vmovapd 128(%[b]), %%ymm14\n\t"
"vfmadd231pd %%ymm14, %%ymm12, %%ymm0\n\t"
"vfmadd231pd %%ymm14, %%ymm13, %%ymm6\n\t"
"vmovapd 160(%[b]), %%ymm14\n\t"
"vfmadd231pd %%ymm14, %%ymm12, %%ymm1\n\t"
"vfmadd231pd %%ymm14, %%ymm13, %%ymm7\n\t"
"vmovapd 128(%[b1]), %%ymm14\n\t"
"vfmadd231pd %%ymm14, %%ymm12, %%ymm2\n\t"
"vfmadd231pd %%ymm14, %%ymm13, %%ymm8\n\t"
"vmovapd 160(%[b1]), %%ymm14\n\t"
"vfmadd231pd %%ymm14, %%ymm12, %%ymm3\n\t"
"vfmadd231pd %%ymm14, %%ymm13, %%ymm9\n\t"
"vmovapd 128(%[b2]), %%ymm14\n\t"
"vfmadd231pd %%ymm14, %%ymm12, %%ymm4\n\t"
"vfmadd231pd %%ymm14, %%ymm13, %%ymm10\n\t"
"vmovapd 160(%[b2]), %%ymm14\n\t"
"vfmadd231pd %%ymm14, %%ymm12, %%ymm5\n\t"
"vfmadd231pd %%ymm14, %%ymm13, %%ymm11\n\t"
"prefetcht0 704(%[b])\n\t"
"prefetcht0 704(%[b1])\n\t"
"prefetcht0 704(%[b2])\n\t"
"vbroadcastsd %c[o0]+24(%[a]), %%ymm12\n\t"
"vbroadcastsd %c[o1]+24(%[a]), %%ymm13\n\t"
"vmovapd 192(%[b]), %%ymm14\n\t"
"vfmadd231pd %%ymm14, %%ymm12, %%ymm0\n\t"
"vfmadd231pd %%ymm14, %%ymm13, %%ymm6\n\t"
"vmovapd 224(%[b]), %%ymm14\n\t"
"vfmadd231pd %%ymm14, %%ymm12, %%ymm1\n\t"
"vfmadd231pd %%ymm14, %%ymm13, %%ymm7\n\t"
"vmovapd 192(%[b1]), %%ymm14\n\t"
"vfmadd231pd %%ymm14, %%ymm12, %%ymm2\n\t"
"vfmadd231pd %%ymm14, %%ymm13, %%ymm8\n\t"
"vmovapd 224(%[b1]), %%ymm14\n\t"
"vfmadd231pd %%ymm14, %%ymm12, %%ymm3\n\t"
"vfmadd231pd %%ymm14, %%ymm13, %%ymm9\n\t"
"vmovapd 192(%[b2]), %%ymm14\n\t"
"vfmadd231pd %%ymm14, %%ymm12, %%ymm4\n\t"
"vfmadd231pd %%ymm14, %%ymm13, %%ymm10\n\t"
"vmovapd 224(%[b2]), %%ymm14\n\t"
"vfmadd231pd %%ymm14, %%ymm12, %%ymm5\n\t"
"vfmadd231pd %%ymm14, %%ymm13, %%ymm11\n\t"
"add $32, %[a]\n\t"
"add $256, %[b]\n\t"
"add $256, %[b1]\n\t"
"add $256, %[b2]\n\t"
"sub $4, %[kk]\n\t"
"jnz 1b\n\t"
"vmovntpd %%ymm0, 0(%[out])\n\t"
"vmovntpd %%ymm1, 32(%[out])\n\t"
"vmovntpd %%ymm2, 64(%[out])\n\t"
"vmovntpd %%ymm3, 96(%[out])\n\t"
"vmovntpd %%ymm4, 128(%[out])\n\t"
"vmovntpd %%ymm5, 160(%[out])\n\t"
"add %[stride], %[out]\n\t"
"vmovntpd %%ymm6, 0(%[out])\n\t"
"vmovntpd %%ymm7, 32(%[out])\n\t"
"vmovntpd %%ymm8, 64(%[out])\n\t"
"vmovntpd %%ymm9, 96(%[out])\n\t"
"vmovntpd %%ymm10, 128(%[out])\n\t"
"vmovntpd %%ymm11, 160(%[out])\n\t"
: [a] "+&r"(a),[b] "+&r"(b),[b1] "+&r"(b1),[b2] "+&r"(b2),[out] "+&r"(out),[kk] "+&r"(kk)
: [stride] "r"(stride),[o0] "i"(0),[o1] "i"((BASE+24)*8)
: "memory","cc","ymm0","ymm1","ymm2","ymm3","ymm4","ymm5","ymm6","ymm7","ymm8","ymm9","ymm10","ymm11","ymm12","ymm13","ymm14","ymm15");
}
struct RootPending { double*c;int ldc;double*m;int pos;bool active;double*m5;double*m6; };
static RootPending root_pending={nullptr,0,nullptr,0,false};
static constexpr int ROOT_STEP=32;
static inline void root_prefetch() {
if(!root_pending.active)return;
const double*b0=root_pending.c+(size_t)((root_pending.pos+0)>>11)*root_pending.ldc+((root_pending.pos+0)&2047);
const double*b1=root_pending.c+(size_t)((root_pending.pos+0)>>11)*root_pending.ldc+((root_pending.pos+0)&2047)+(size_t)2048*root_pending.ldc;
const double*b2=root_pending.c+(size_t)((root_pending.pos+0)>>11)*root_pending.ldc+((root_pending.pos+0)&2047)+(size_t)2048*root_pending.ldc+2048;
const double*m=root_pending.m+root_pending.pos+32,*m5=root_pending.m5+root_pending.pos+32,*m6=root_pending.m6+root_pending.pos+32;
for(int t=0;t<ROOT_STEP;t+=8){
_mm_prefetch((const char*)(b0+t),_MM_HINT_T1);
_mm_prefetch((const char*)(b1+t),_MM_HINT_T1);
_mm_prefetch((const char*)(b2+t),_MM_HINT_T1);
_mm_prefetch((const char*)(m+t),_MM_HINT_T1);
_mm_prefetch((const char*)(m5+t),_MM_HINT_T1);
_mm_prefetch((const char*)(m6+t),_MM_HINT_T1);
}
}
static inline void root_step() {
int i=root_pending.pos>>11,j=root_pending.pos&2047;
for(int t=0;t<ROOT_STEP;t+=4){
double*c11=root_pending.c+(size_t)i*root_pending.ldc+j+t;
double*c21=c11+(size_t)2048*root_pending.ldc,*c22=c21+2048;
double*m=root_pending.m+root_pending.pos+t;
__m256d m1=_mm256_load_pd(c11),m2=_mm256_load_pd(c21),m4=_mm256_load_pd(c22);
__m256d m5=_mm256_load_pd(m),m6=_mm256_load_pd(root_pending.m5+root_pending.pos+t),m7=_mm256_load_pd(root_pending.m6+root_pending.pos+t);
__m256d u2=_mm256_add_pd(m1,m6),u3=_mm256_add_pd(u2,m7),u4=_mm256_add_pd(u2,m5);
_mm256_store_pd(c11,_mm256_add_pd(m1,m2));
_mm256_store_pd(c21,_mm256_sub_pd(u3,m4));
_mm256_store_pd(c22,_mm256_add_pd(u3,m5));
_mm256_store_pd(m,u4);
}
root_pending.pos+=ROOT_STEP;
if(root_pending.pos==2048*2048)root_pending.active=false;
}
static void root_finish(){while(root_pending.active)root_step();}
static bool r12_arm=false;
static int r12_r=0,r12_ph=0,r12_off=0;
static void r12_emit(void);
static void r12_trail_body(void);
static inline void r12_trail(void){ if(r12_arm) r12_trail_body(); }
static void pend2_partial_step(void) {
if(pending.active)return;          
if(root_pending.active)return;     
int i=pend2.pos>>10, j=pend2.pos&1023;
for(int t=0;t<64;t+=4) {
double *c11=pend2.c+(size_t)i*pend2.ldc+j+t,*c21=c11+(size_t)1024*pend2.ldc,*c22=c21+1024;
const double *mm=pend2.m+pend2.pos+t;
const double *pp=pend2.add+(size_t)i*2048+j+t;
__m256d m1=_mm256_load_pd(c11),m2=_mm256_load_pd(c21),m4=_mm256_load_pd(c22);
__m256d m5=_mm256_load_pd(mm),m6=_mm256_load_pd(mm+(1024*1024+UPPER_PAGE_PAD)),
m7=_mm256_load_pd(mm+2*(1024*1024+UPPER_PAGE_PAD));
__m256d p11=_mm256_load_pd(pp),p21=_mm256_load_pd(pp+(size_t)1024*2048),
p22=_mm256_load_pd(pp+(size_t)1024*2048+1024);
__m256d u2=_mm256_add_pd(m1,m6),u3=_mm256_add_pd(u2,m7),u4=_mm256_add_pd(u2,m5);
__m256d r11=_mm256_add_pd(m1,m2),r21=_mm256_sub_pd(u3,m4),r22=_mm256_add_pd(u3,m5);
r11=_mm256_add_pd(p11,r11);                 
r21=_mm256_add_pd(p21,r21);
r22=_mm256_add_pd(p22,r22);
_mm256_store_pd(c11,r11);_mm256_store_pd(c21,r21);_mm256_store_pd(c22,r22);
_mm256_store_pd((double*)pend2.m+pend2.pos+t,u4);   
}
pend2.pos+=64;
if(pend2.pos>=1024*1024)pend2.active=false;
}
static void pend2_partial_finish(void) {
if(pending.active)pending_finish();
if(root_pending.active)root_finish();
while(pend2.active)pend2_partial_step();
for(int i=0;i<0;i++){
double *c12=pend2.c+(size_t)i*pend2.ldc+1024;
const double *mm=pend2.m+(size_t)i*1024;
const double *pp=pend2.add+(size_t)i*2048+1024;
for(int j=0;j<1024;j+=16){
for(int t=0;t<16;t+=4){
__m256d u4=_mm256_load_pd(mm+j+t),m3=_mm256_load_pd(c12+j+t),p12=_mm256_load_pd(pp+j+t);
_mm256_store_pd(c12+j+t,_mm256_add_pd(p12,_mm256_add_pd(u4,m3)));
}
if(j+32<1024){
_mm_prefetch((const char*)(mm+j+32),_MM_HINT_T0);
_mm_prefetch((const char*)(c12+j+32),_MM_HINT_T0);
_mm_prefetch((const char*)(pp+j+32),_MM_HINT_T0);
}
if(i+1<1024)_mm_prefetch((const char*)(pend2.add+(size_t)(i+1)*2048+1024),_MM_HINT_T0);
}
}
while(r12_r<512)r12_emit();
pend2.active=false;
}
static const int lperm[7]={4,0,1,3,2,5,6};
static int g_fuse_sd[7], g_fuse_sgn[7];
static int g_fuse_sd_cur=0, g_fuse_sgn_cur=0;
static const int d1a2_bi[7]={0,0,1,2,3,0,2}, d1a2_bj[7]={3,-1,3,0,-1,1,3}, d1a2_bs[7]={1,0,-1,-1,0,1,1};
static void d1a2_fuse_tab(void){
for(int r=0;r<7;r++){g_fuse_sd[r]=0;g_fuse_sgn[r]=0;}
for(int r=1;r<7;r++){
int q=lperm[r], pq=lperm[r-1];
if(d1a2_bs[pq]!=0) continue;
if(d1a2_bj[q]==d1a2_bi[pq]){ g_fuse_sd[r]=1; g_fuse_sgn[r]=d1a2_bs[q]; }
else if(d1a2_bi[q]==d1a2_bi[pq]){ g_fuse_sd[r]=2; g_fuse_sgn[r]=d1a2_bs[q]; }
}
}
__attribute__((noinline)) static void pk_fuse(int n,Mat b,double *BP,int sgn,int side){
const double *sx=b.x; const size_t sl=b.ld;
for(int k=0;k<n;k+=2)for(int j=0;j<n;j+=8) {
double *p=BP+(size_t)j*(n+1)+k*8;
__m256d q0=_mm256_load_pd(p+0),q1=_mm256_load_pd(p+4);
__m256d q2=_mm256_load_pd(p+8),q3=_mm256_load_pd(p+12);
__m256d s0=_mm256_load_pd(sx+(size_t)(k+0)*sl+j),   s1=_mm256_load_pd(sx+(size_t)(k+0)*sl+j+4);
__m256d s2=_mm256_load_pd(sx+(size_t)(k+1)*sl+j),   s3=_mm256_load_pd(sx+(size_t)(k+1)*sl+j+4);
if(side==1){
if(sgn>0){q0=_mm256_add_pd(s0,q0);q1=_mm256_add_pd(s1,q1);q2=_mm256_add_pd(s2,q2);q3=_mm256_add_pd(s3,q3);}
else     {q0=_mm256_sub_pd(s0,q0);q1=_mm256_sub_pd(s1,q1);q2=_mm256_sub_pd(s2,q2);q3=_mm256_sub_pd(s3,q3);}
} else {
if(sgn>0){q0=_mm256_add_pd(q0,s0);q1=_mm256_add_pd(q1,s1);q2=_mm256_add_pd(q2,s2);q3=_mm256_add_pd(q3,s3);}
else     {q0=_mm256_sub_pd(q0,s0);q1=_mm256_sub_pd(q1,s1);q2=_mm256_sub_pd(q2,s2);q3=_mm256_sub_pd(q3,s3);}
}
_mm256_store_pd(p+0,q0);_mm256_store_pd(p+4,q1);
_mm256_store_pd(p+8,q2);_mm256_store_pd(p+12,q3);
}
}
static void d1a2_pk(int n,Mat b,double *BP){
if(g_fuse_sd_cur) pk_fuse(n,b,BP,g_fuse_sgn_cur,g_fuse_sd_cur);
else for(int k=0;k<n;k+=2)for(int j=0;j<n;j+=8) {
double *p=BP+(size_t)j*(n+1)+k*8;
_mm256_store_pd(p+0,ld(b,k+0,j));_mm256_store_pd(p+4,ld(b,k+0,j+4));
_mm256_store_pd(p+8,ld(b,k+1,j));_mm256_store_pd(p+12,ld(b,k+1,j+4));
}
}
static void pre(int h,Mat a,double *s1,double *s2,double *s4,double *s3,int ld3,bool right);
struct D4ahPP { const double *s1; int l1; const double *s2; int l2; double *d; int l3; int sg; };
struct D4ahPT { int h; int i, j; int pass; D4ahPP p[4]; };
static D4ahPT d4ah_pt[8];
static int d4ah_ptn=0;
static int d4ah_on=1;
static long long d4ah_pchunks=0;
alignas(32) static double d4ah_scr[8][256];
static const double *d4ai_L0,*d4ai_L1,*d4ai_L2; static double *d4ai_D0,*d4ai_D1;static double *d4ai_SC;
static int d4ai_mode=0, d4ai_side=0;
#ifndef D4AJ_R
#define D4AJ_R 2
#endif
static int d4aj_skip=0;
static void d4ah_push(int h,const D4ahPP *q){          
if(d4ah_ptn>=8) return;
D4ahPT *t=&d4ah_pt[d4ah_ptn++];
t->h=h; t->i=0; t->j=0; t->pass=0;
for(int k=0;k<4;k++) t->p[k]=q[k];
}
static void d4ah_run(D4ahPT *t){                        
for(int i=t->i;i<t->h;i++){
for(int ps=(i==t->i?t->pass:0); ps<4; ps++){
D4ahPP *p=&t->p[ps];
const double *r1=p->s1+(size_t)i*p->l1, *r2=p->s2+(size_t)i*p->l2;
double *rr=p->d+(size_t)i*p->l3;
for(int j=0;j<t->h;j+=4){
__m256d u=_mm256_loadu_pd(r1+j), v=_mm256_loadu_pd(r2+j);
_mm256_storeu_pd(rr+j, p->sg>0?_mm256_add_pd(u,v):_mm256_sub_pd(u,v));
}
}
}
t->i=t->h;
}
static int d4ah_pre_arm(int h,Mat a,double *s1,double *s2,double *s4,double *s3,int ld3,bool right){
int n0=d4ah_ptn;
if(!d4ah_on||a.sign||h<256||(h&255)) { pre(h,a,s1,s2,s4,s3,ld3,right); return n0; }
const double *x=a.x; int ld=a.ld;
D4ahPP q[4];
if(right){   
q[0]=(D4ahPP){x+h,ld, x,ld, s1,h, -1};
q[1]=(D4ahPP){x+(size_t)h*ld+h,ld, s1,h, s2,h, -1};
q[2]=(D4ahPP){s2,h, x+(size_t)h*ld,ld, s4,h, -1};
q[3]=(D4ahPP){x+(size_t)h*ld+h,ld, x+h,ld, s3,ld3, -1};
} else {     
q[0]=(D4ahPP){x+(size_t)h*ld,ld, x+(size_t)h*ld+h,ld, s1,h, +1};
q[1]=(D4ahPP){s1,h, x,ld, s2,h, -1};
q[2]=(D4ahPP){x+h,ld, s2,h, s4,h, -1};
q[3]=(D4ahPP){x,ld, x+(size_t)h*ld,ld, s3,ld3, -1};
}
d4ah_push(h,q);
return n0;
}
static inline int d4ah_pre_chunk(const double **q1,const double **q2,double **qd,int *qsg){
if(!d4ah_ptn) return 0;
D4ahPT *t=&d4ah_pt[d4ah_ptn-1];
if(t->h>=2048){                 
if(d4aj_skip>0){ d4aj_skip--; return 0; }
d4aj_skip=D4AJ_R-1;
}
D4ahPP *p=t->p;
size_t b0=(size_t)t->i*p[0].l3+t->j, b1=(size_t)t->i*p[1].l3+t->j, b2=(size_t)t->i*p[2].l3+t->j, b3=(size_t)t->i*p[3].l3+t->j;
int side=(p[0].sg>0)?0:1;
double *SCR=d4ah_scr[t-&d4ah_pt[0]];
if(t->pass==0){
d4ai_L0=p[0].s1+(size_t)t->i*p[0].l1+t->j;
d4ai_L1=p[0].s2+(size_t)t->i*p[0].l2+t->j;
d4ai_L2=(side==0? p[1].s2+(size_t)t->i*p[1].l2 : p[1].s1+(size_t)t->i*p[1].l1)+t->j;
d4ai_D0=p[0].d+b0; d4ai_D1=p[3].d+b3; d4ai_SC=SCR;
} else {
if(side==0){ d4ai_L0=SCR; d4ai_L1=p[1].s2+(size_t)t->i*p[1].l2+t->j; }
else       { d4ai_L0=p[1].s1+(size_t)t->i*p[1].l1+t->j; d4ai_L1=SCR; }
d4ai_L2=(side==0? p[2].s1+(size_t)t->i*p[2].l1 : p[2].s2+(size_t)t->i*p[2].l2)+t->j;
d4ai_D0=p[1].d+b1; d4ai_D1=p[2].d+b2;
}
d4ai_mode=t->pass; d4ai_side=side;
*q1=d4ai_L0; *q2=d4ai_L1; *qd=(double*)d4ai_L2; *qsg=side;
d4ah_pchunks++;
if(++t->pass>=2){ t->pass=0; t->j+=256; if(t->j>=t->h){ t->j=0; if(++t->i>=t->h) d4ah_ptn--; } }
return 1;
}
static void d4ah_pre_fin_to(int k){
while(d4ah_ptn>k){ D4ahPT *t=&d4ah_pt[d4ah_ptn-1]; d4ah_run(t); d4ah_ptn--; }
_mm_sfence();   
}
__attribute__((noinline)) static void d4bc_carry(int n,const double *a,const double *b,double *c,int ldc,int mr,
const double *pks,const double *pky,double *pkd,int sg,int sx,int sy){
if(sg>0) kernel< 1>(n,a,b,c,ldc,mr,pks,pky,pkd,sx,sy);
else if(sg<0) kernel<-1>(n,a,b,c,ldc,mr,pks,pky,pkd,sx,sy);
else kernel< 2>(n,a,b,c,ldc,mr,pks,pky,pkd,sx,sy);
}
template<int TASKS> static void leaf_impl(int leaf_n,Mat a,Mat b,double *c,int ldc,const Mat *anext=nullptr,bool odd=false,bool needp0=true) {
constexpr int n=BASE;
static_assert(BASE==512 && MC==30 && MR==6,"exact tail geometry");
d1a2_pk(n,b,bp);
if constexpr(TASKS&1)pending_prefetch();
double *pkb[2]={ap,ap2};
int cur=(odd?1:0);
if(needp0) {   
int mc0=n<MC?n:MC;
for(int r=0;r<mc0;r+=MR) {
int rows=mc0-r<MR?mc0-r:MR;
for(int i=0;i<rows;i++)for(int k=0;k<n;k+=4)
_mm256_store_pd(pkb[cur]+(size_t)(r+i)*(BASE+24)+k,ld(a,r+i,k));
for(int i=rows;i<MR;i++)for(int k=0;k<n;k+=4)
_mm256_store_pd(pkb[cur]+(size_t)(r+i)*(BASE+24)+k,_mm256_setzero_pd());
}
}
for(int ic=0;ic<n;) {
int mc=n-ic<=MC+MR?n-ic:MC;
if(mc==MC+2) {
const int d4bc_nch2=anext?30*2:0;
int d4bc_ck=0, d4bc_skip=(n/NR)*((MC+MR-1)/MR)-d4bc_nch2;
const int d4bc_qd=n/2;
for(int j=0;j<n;j+=24) {
for(int jj=j;jj<j+24 && jj<n;jj+=NR)for(int ri=0;ri<MC;ri+=MR) {
const int r=(jj&NR)?MC-MR-ri:ri;
if(d4bc_nch2 && d4bc_ck<d4bc_nch2 && --d4bc_skip<0) {
const int row=d4bc_ck/2, off=(d4bc_ck%2)*d4bc_qd;
const double *pks=anext->x+(size_t)row*anext->ld+(size_t)off;
const double *pky=anext->sign?(anext->y+(size_t)row*anext->ly+(size_t)off):pk_zero;
double *pkd=pkb[cur^1]+(size_t)row*(BASE+24)+(size_t)off;
d4bc_ck++;
d4bc_carry(n,pkb[cur]+(size_t)r*(BASE+24),bp+(size_t)jj*(n+1),c+(size_t)(ic+r)*ldc+jj,ldc,MR,pks,pky,pkd,anext->sign,anext->ld,anext->ly);
} else
{ if(jj!=0 && (ri&2)) kernel_reverse(n,pkb[cur]+(size_t)r*(BASE+24),bp+(size_t)jj*(n+1),c+(size_t)(ic+r)*ldc+jj,ldc,MR); else kernel_plain(n,pkb[cur]+(size_t)r*(BASE+24),bp+(size_t)jj*(n+1),c+(size_t)(ic+r)*ldc+jj,ldc,MR); }
if((TASKS&1) && pending.active && (++pending_tick%PERIOD)==0 && (!pending.prog || pending.pos<BASE*pending.rowlim)) {pending_step_unchecked();if(pending.prog)pending_step_unchecked();pending_prefetch();r12_trail();}
if((TASKS&2) && pend2.active) { if(pend2.part)pend2_partial_step(); else pend2_step(); pend2_prefetch();}
if((TASKS&4) && root_pending.active && !pending.active) {root_step();root_prefetch();}
}
if(j+24<=n)kernel_tail24(n,pkb[cur]+(size_t)MC*(BASE+24),bp+(size_t)j*(n+1),c+(size_t)(ic+MC)*ldc+j,ldc);
else kernel2(n,pkb[cur]+(size_t)MC*(BASE+24),bp+(size_t)j*(n+1),c+(size_t)(ic+MC)*ldc+j,ldc);
}
ic+=mc;
if((TASKS&1) && pending.prog)pending.rowlim=ic;
continue;
}
const int ic2=ic+mc;
const int nxt=(ic2<n)?(n-ic2<=MC+MR?n-ic2:MC):0;
const int rows2=nxt; 
const int nch=rows2*2;                     
const int qd=n/2;                         
static_assert(PERIOD==1,"one reconstruction opportunity per microtile");
const bool is_producer=pending.prog;
const int merge_limit=is_producer?BASE*pending.rowlim:BASE*BASE;
int merge_left=(TASKS&1)&&pending.active&&merge_limit>pending.pos?(merge_limit-pending.pos)/(STEP*(is_producer?2:1)):0;
int ck=0; int skip=(n/NR)*((mc+MR-1)/MR)-nch;
for (int j=0;j<n;j+=NR)for(int ri=0;ri<mc;ri+=MR) {
const int r=(j&NR)?mc-MR-ri:ri;
if(mc-r==2) {   
kernel2(n,pkb[cur]+(size_t)r*(BASE+24),bp+(size_t)j*(n+1),c+(size_t)(ic+r)*ldc+j,ldc);
continue;
}
const double *pks=pk_zero,*pky=pk_zero,*pkm=pk_mp;
double *pkd=pk_zero;
int pkc=0; int psg=0;
if(ck<nch && --skip<0) {
const int row=ck/2, off=(ck%2)*qd;
if(row<nxt) {
pks=a.x+(size_t)(ic2+row)*a.ld+(size_t)off;
pky=a.sign?(a.y+(size_t)(ic2+row)*a.ly+(size_t)off):pk_zero;
pkm=a.sign>0?pk_mp:pk_mn;
}
pkd=pkb[cur^1]+(size_t)row*(BASE+24)+(size_t)off;
pkc=8;
ck++;
}
if(!pkc && d4ah_ptn){
if(d4ah_pre_chunk(&pks,&pky,&pkd,&psg)) pkc=9;
}
if(pkc==9){
const double *PL0=d4ai_L0,*PL1=d4ai_L1,*PL2=d4ai_L2; double *PD0=d4ai_D0,*PD1=d4ai_D1;
if(d4ai_mode==0){ double *PSC=d4ai_SC; if(d4ai_side==0) kpA< 1>(n,pkb[cur]+(size_t)r*(BASE+24),bp+(size_t)j*(n+1),c+(size_t)(ic+r)*ldc+j,ldc,mc-r<MR?mc-r:MR,PL0,PL1,PL2,PD0,PD1,PSC);
else             kpA<-1>(n,pkb[cur]+(size_t)r*(BASE+24),bp+(size_t)j*(n+1),c+(size_t)(ic+r)*ldc+j,ldc,mc-r<MR?mc-r:MR,PL0,PL1,PL2,PD0,PD1,PSC); }
else            { if(d4ai_side==0) kpB< 0>(n,pkb[cur]+(size_t)r*(BASE+24),bp+(size_t)j*(n+1),c+(size_t)(ic+r)*ldc+j,ldc,mc-r<MR?mc-r:MR,PL0,PL1,PL2,PD0,PD1);
else             kpB< 1>(n,pkb[cur]+(size_t)r*(BASE+24),bp+(size_t)j*(n+1),c+(size_t)(ic+r)*ldc+j,ldc,mc-r<MR?mc-r:MR,PL0,PL1,PL2,PD0,PD1); }
}
else if(!pkc) { if(j!=0 && (ri&2)) kernel_reverse(n,pkb[cur]+(size_t)r*(BASE+24),bp+(size_t)j*(n+1),c+(size_t)(ic+r)*ldc+j,ldc,mc-r<MR?mc-r:MR); else kernel_plain(n,pkb[cur]+(size_t)r*(BASE+24),bp+(size_t)j*(n+1),c+(size_t)(ic+r)*ldc+j,ldc,mc-r<MR?mc-r:MR); }
else if(psg>0) kernel<1>(n,pkb[cur]+(size_t)r*(BASE+24),bp+(size_t)j*(n+1),c+(size_t)(ic+r)*ldc+j,ldc,mc-r<MR?mc-r:MR,pks,pky,pkd,a.ld,a.ly);
else if(psg<0) kernel<-1>(n,pkb[cur]+(size_t)r*(BASE+24),bp+(size_t)j*(n+1),c+(size_t)(ic+r)*ldc+j,ldc,mc-r<MR?mc-r:MR,pks,pky,pkd,a.ld,a.ly);
else if(a.sign==0) kernel<2>(n,pkb[cur]+(size_t)r*(BASE+24),bp+(size_t)j*(n+1),c+(size_t)(ic+r)*ldc+j,ldc,mc-r<MR?mc-r:MR,pks,pky,pkd,a.ld,a.ly);
else if(a.sign>0) kernel<1>(n,pkb[cur]+(size_t)r*(BASE+24),bp+(size_t)j*(n+1),c+(size_t)(ic+r)*ldc+j,ldc,mc-r<MR?mc-r:MR,pks,pky,pkd,a.ld,a.ly);
else kernel<-1>(n,pkb[cur]+(size_t)r*(BASE+24),bp+(size_t)j*(n+1),c+(size_t)(ic+r)*ldc+j,ldc,mc-r<MR?mc-r:MR,pks,pky,pkd,a.ld,a.ly);
if((TASKS&1) && merge_left>0){
--merge_left;pending_step_unchecked();if(is_producer)pending_step_unchecked();pending_prefetch();r12_trail();
}
if((TASKS&2) && pend2.active) { if(pend2.part)pend2_partial_step(); else pend2_step(); pend2_prefetch(); }
if((TASKS&4) && root_pending.active && !pending.active) {root_step();root_prefetch();}
}
for(int cc=ck;cc<nch;cc++) {
const int row=cc/2, off=(cc%2)*qd;
if(row<nxt) {
for(int k=off;k<off+qd;k+=4)
_mm256_store_pd(pkb[cur^1]+(size_t)row*(BASE+24)+k, ld(a,ic2+row,k));
} else {
for(int k=off;k<off+qd;k+=4)
_mm256_store_pd(pkb[cur^1]+(size_t)row*(BASE+24)+k, _mm256_setzero_pd());
}
}
cur^=1;
ic+=mc;
if((TASKS&1) && pending.prog)pending.rowlim=ic;
}
if(n>=NTMIN)_mm_sfence();
}
static void leaf(int n,Mat a,Mat b,double*c,int ldc,const Mat *anext=nullptr,bool odd=false,bool needp0=true){int mode=pending.active+2*pend2.active+4*root_pending.active;switch(mode){case 7:leaf_impl<7>(n,a,b,c,ldc,anext,odd,needp0);break;case 0:leaf_impl<0>(n,a,b,c,ldc,anext,odd,needp0);break;case 1:leaf_impl<1>(n,a,b,c,ldc,anext,odd,needp0);break;case 2:leaf_impl<2>(n,a,b,c,ldc,anext,odd,needp0);break;case 3:leaf_impl<3>(n,a,b,c,ldc,anext,odd,needp0);break;case 4:leaf_impl<4>(n,a,b,c,ldc,anext,odd,needp0);break;case 5:leaf_impl<5>(n,a,b,c,ldc,anext,odd,needp0);break;case 6:leaf_impl<6>(n,a,b,c,ldc,anext,odd,needp0);break;}}
static void sum(int h,double *dst,Mat a,Mat b,int sign) {
for(int i=0;i<h;i++)for(int j=0;j<h;j+=8) {
__m256d v=sign>0?_mm256_add_pd(ld(a,i,j),ld(b,i,j)):_mm256_sub_pd(ld(a,i,j),ld(b,i,j));
__m256d w=sign>0?_mm256_add_pd(ld(a,i,j+4),ld(b,i,j+4)):_mm256_sub_pd(ld(a,i,j+4),ld(b,i,j+4));
asm volatile("" : "+x"(v), "+x"(w) : : "memory");
if(h>=SUMNT) {
_mm256_stream_pd(dst+(size_t)i*h+j,v);_mm256_stream_pd(dst+(size_t)i*h+j+4,w);
} else {
_mm256_store_pd(dst+(size_t)i*h+j,v);_mm256_store_pd(dst+(size_t)i*h+j+4,w);
}
}
if(h>=SUMNT)_mm_sfence();
}
static void combine(int h,double *c,int ldc,const double *m,bool wino=false) {
size_t z=(size_t)h*h+(wino?UPPER_PAGE_PAD:0);
for(int i=0;i<h;i++)for(int j=0;j<h;j+=4) {
size_t t=(size_t)i*h+j;
double *c11=c+(size_t)i*ldc+j,*c12=c11+h,*c21=c+(size_t)(i+h)*ldc+j,*c22=c21+h;
__m256d m1=_mm256_load_pd(c11),m2=_mm256_load_pd(c21),m3=_mm256_load_pd(c12),m4=_mm256_load_pd(c22),
m5=_mm256_load_pd(m+t),m6=_mm256_load_pd(m+z+t),m7=_mm256_load_pd(m+2*z+t);
__m256d r11,r12,r21,r22;
if(wino) {
__m256d u2=_mm256_add_pd(m1,m6),u3=_mm256_add_pd(u2,m7),u4=_mm256_add_pd(u2,m5);
r11=_mm256_add_pd(m1,m2);r12=_mm256_add_pd(u4,m3);r21=_mm256_sub_pd(u3,m4);r22=_mm256_add_pd(u3,m5);
} else {
r11=_mm256_add_pd(_mm256_sub_pd(_mm256_add_pd(m1,m4),m5),m7);r12=_mm256_add_pd(m3,m5);r21=_mm256_add_pd(m2,m4);
r22=_mm256_add_pd(_mm256_add_pd(_mm256_sub_pd(m1,m2),m3),m6);
}
_mm256_store_pd(c11,r11);_mm256_store_pd(c12,r12);_mm256_store_pd(c21,r21);_mm256_store_pd(c22,r22);
}
}
static inline __m256d raw_pre_ld(Mat a,int i,int j){return _mm256_load_pd(a.x+(size_t)i*a.ld+j);}
static void pre(int h,Mat a,double *s1,double *s2,double *s4,double *s3,int ld3,bool right) {
for(int i=0;i<h;i++)for(int j=0;j<h;j+=8) {
__m256d a0=raw_pre_ld(a,i,j),a1=raw_pre_ld(a,i,j+4),b0=raw_pre_ld(a,i,j+h),b1=raw_pre_ld(a,i,j+h+4),
c0=raw_pre_ld(a,i+h,j),c1=raw_pre_ld(a,i+h,j+4),d0=raw_pre_ld(a,i+h,j+h),d1=raw_pre_ld(a,i+h,j+h+4);
__m256d p0,p1,q0,q1,r0,r1,t0,t1;
if(!right) {
p0=_mm256_add_pd(c0,d0);p1=_mm256_add_pd(c1,d1);
q0=_mm256_sub_pd(p0,a0);q1=_mm256_sub_pd(p1,a1);
r0=_mm256_sub_pd(b0,q0);r1=_mm256_sub_pd(b1,q1);
t0=_mm256_sub_pd(a0,c0);t1=_mm256_sub_pd(a1,c1);
}else {
p0=_mm256_sub_pd(b0,a0);p1=_mm256_sub_pd(b1,a1);
q0=_mm256_sub_pd(d0,p0);q1=_mm256_sub_pd(d1,p1);
r0=_mm256_sub_pd(q0,c0);r1=_mm256_sub_pd(q1,c1);
t0=_mm256_sub_pd(d0,b0);t1=_mm256_sub_pd(d1,b1);
}
size_t off=(size_t)i*h+j,off3=(size_t)i*ld3+j;
_mm256_stream_pd(s1+off,p0);_mm256_stream_pd(s1+off+4,p1);
_mm256_stream_pd(s2+off,q0);_mm256_stream_pd(s2+off+4,q1);
_mm256_stream_pd(s4+off,r0);_mm256_stream_pd(s4+off+4,r1);
_mm256_stream_pd(s3+off3,t0);_mm256_stream_pd(s3+off3+4,t1);
}
_mm_sfence();
}
template<int N> static void rec(int rec_n,Mat a,Mat b,double *c,int ldc,double *work,bool root_last=false,const Mat *d4bd_succ=nullptr,bool d4bd_cin=false,bool d4be_r12src=false) {
constexpr int n=N;
if constexpr(n<=BASE){leaf(n,a,b,c,ldc);return;} else {
constexpr int h=n/2;size_t z=(size_t)h*h;
Mat aa[4]={{a.x,a.ld},{a.x+h,a.ld},{a.x+(size_t)h*a.ld,a.ld},{a.x+(size_t)h*a.ld+h,a.ld}};
Mat bb[4]={{b.x,b.ld},{b.x+h,b.ld},{b.x+(size_t)h*b.ld,b.ld},{b.x+(size_t)h*b.ld+h,b.ld}};
double *m=n==2*BASE?work+LEAF_PAD:work;
if(n==2*BASE && pending.active && pending.m==m)m=pending_work+LEAF_PAD;
size_t pz=z+(n==2*BASE?LEAF_PAD:UPPER_PAGE_PAD);
double *p=work+3*z,*q=p+z,*sub=q+z;
double *out[7]={c,c+(size_t)h*ldc,c+h,c+(size_t)h*ldc+h,m,m+pz,m+2*pz};
if(n>BASE*2) {
pending_finish();
double *pa0=m+3*pz,*pa1=pa0+pz,*pa2=pa1+pz,*pb0=pa2+pz,*pb1=pb0+pz,*pb2=pb1+pz;
sub=pb2+pz;
if constexpr (N==4096) { out[4]=pb2; out[5]=pa0; out[6]=pa1; }
int d4ah_n0=d4ah_ptn;
const bool d4bb_out=(ldc==2*h); const int d4bb_l3=d4bb_out? h : ldc;
d4ah_pre_arm(h,a,pa0,pa1,pa2,c,d4bb_l3,false);   
d4ah_pre_arm(h,b,pb0,pb1,pb2,d4bb_out? c+(size_t)h*h : c+h,d4bb_l3,true);  
Mat d4bd_d[5];
d4bd_d[0]=d4bd_first(aa[3],h);                 
d4bd_d[1]=d4bd_first(Mat{pa1,h},h);            
{ Mat d4bd_c5=d4bb_out?Mat{c,h}:Mat{c,ldc}; d4bd_d[2]=d4bd_first(d4bd_c5,h); }   
d4bd_d[3]=d4bd_first(aa[0],h);                 
d4bd_d[4]=d4bd_first(Mat{pa2,h},h);            
rec<h>(h,aa[1],bb[2],out[1],ldc,sub,false,&d4bd_d[0],d4bd_cin);
d4ah_pre_fin_to(d4ah_n0+1);
rec<h>(h,aa[3],{pb2,h},out[3],ldc,sub,false,nullptr,true);   
if(root_last) {
pend2_finish(); 
root_pending.active=true;
root_prefetch();
}
#if !B2M_NO_BARA
pend2_finish();            
#endif
d4ah_pre_fin_to(d4ah_n0);
rec<h>(h,{pa0,h},{pb0,h},out[4],h,sub,false,&d4bd_d[1],false);
rec<h>(h,{pa1,h},{pb1,h},out[5],h,sub,false,&d4bd_d[2],true);
if(d4bb_out) rec<h>(h,{c,h},{c+(size_t)h*h,h},out[6],h,sub,false,&d4bd_d[3],true);   
else rec<h>(h,{c,ldc},{c+h,ldc},out[6],h,sub,false,&d4bd_d[3],true);
rec<h>(h,aa[0],bb[0],out[0],ldc,sub,false,&d4bd_d[4],true);
if(n==4096) {
root_pending={c,ldc,out[4],0,false,out[5],out[6]};
rec<h>(h,{pa2,h},bb[3],out[2],ldc,sub,true,d4bd_succ,true);
pending_finish();root_finish();pend2_finish();
return;
}
if(root_last){ pend2={c,ldc,m,h,(int)pz,0,true}; pend2.add=root_pending.m; pend2.part=true; }
rec<h>(h,{pa2,h},bb[3],out[2],ldc,sub,root_last,d4bd_succ,true,root_last);
pending_finish();
if(root_last){ root_finish(); pend2_partial_finish(); }
else {
pend2_finish();                                   
pend2={c,ldc,m,h,(int)pz,0,true};                 
}
return;
}
const int ai[7]={0,2,0,3,0,2,1},aj[7]={3,3,-1,-1,1,0,3},as[7]={1,1,0,0,1,-1,-1};
const int bi[7]={0,0,1,2,3,0,2},bj[7]={3,-1,3,0,-1,1,3},bs[7]={1,0,-1,-1,0,1,1};
if(h<=BASE) {
Mat xa[7],ya[7];
for(int rr=0;rr<7;rr++){
const int qq=lperm[rr];
xa[rr]=aa[ai[qq]];ya[rr]=bb[bi[qq]];
if(as[qq]){xa[rr].y=aa[aj[qq]].x;xa[rr].ly=a.ld;xa[rr].sign=as[qq];}
if(bs[qq]){ya[rr].y=bb[bj[qq]].x;ya[rr].ly=b.ld;ya[rr].sign=bs[qq];}
}
const bool d4bd_F=((d4bd_np++)&1);   
for(int rr=0;rr<7;rr++){
const int qq=lperm[rr];
g_fuse_sd_cur=g_fuse_sd[rr];g_fuse_sgn_cur=g_fuse_sgn[rr];
Mat x=xa[rr],y=ya[rr];
const Mat *nx=(rr<6)?&xa[rr+1]:d4bd_succ;
const bool oddp=(((rr+d4bd_F)&1)!=0), np0=(rr==0 && !d4bd_cin);
if(rr==6) {
pending_finish();
pending={c,ldc,m,0,true,0,true};
if(d4be_r12src){r12_arm=true;r12_r=0;r12_ph=0;r12_off=0;}
leaf(h,x,y,out[qq],qq<4?ldc:h,nx,oddp,np0);
pending.prog=false;
} else leaf(h,x,y,out[qq],qq<4?ldc:h,nx,oddp,np0);
}
} else {
for(int r=0;r<7;r++) {
Mat x=aa[ai[r]],y=bb[bi[r]];
if(as[r]){sum(h,p,x,aa[aj[r]],as[r]);x={p,h};}
if(bs[r]){sum(h,q,y,bb[bj[r]],bs[r]);y={q,h};}
rec<h>(h,x,y,out[r],r<4?ldc:h,sub);
}
}
if(h==BASE) {
} else {
pending_finish();
combine(h,c,ldc,m);
}
}
}
void matrix_multiply(int n,const double *a,const double *b,double *c) {
d1a2_fuse_tab();
madvise((void*)workspace,sizeof(workspace),MADV_HUGEPAGE);
madvise((void*)pending_work,sizeof(pending_work),MADV_HUGEPAGE);
rec<4096>(n,{a,n},{b,n},c,n,workspace);
d4ah_pre_fin_to(0);
pending_finish();
pend2_finish();
_mm_sfence();
}
static void r12_emit(void){                        
int i=r12_r+(r12_ph?512:0);
double *c12=pend2.c+(size_t)i*pend2.ldc+1024;
const double *mm=pend2.m+(size_t)i*1024;
const double *pp=pend2.add+(size_t)i*2048+1024;
int j0=r12_off,j1=r12_off+128; if(j1>1024)j1=1024;
for(int j=j0;j<j1;j+=16){
for(int t=0;t<16;t+=4){
__m256d u4=_mm256_load_pd(mm+j+t),m3=_mm256_load_pd(c12+j+t),p12=_mm256_load_pd(pp+j+t);
_mm256_store_pd(c12+j+t,_mm256_add_pd(p12,_mm256_add_pd(u4,m3)));
}
if(j+64<1024){
_mm_prefetch((const char*)(mm+j+64),_MM_HINT_T0);
_mm_prefetch((const char*)(c12+j+64),_MM_HINT_T0);
_mm_prefetch((const char*)(pp+j+64),_MM_HINT_T0);
}
}
r12_off=j1;
if(r12_off>=1024){r12_off=0; if(r12_ph){r12_ph=0;r12_r++;} else r12_ph=1;}
}
static void r12_trail_body(void){
for(int rep=0;rep<8;rep++){
if(r12_r>=512)return;
if(pending.pos<(r12_r+1)*BASE)return;
int i=r12_r+(r12_ph?512:0);
int j1=r12_off+128; if(j1>1024)j1=1024;
if(root_pending.pos<=(long long)i*2048+1024+j1-1+64)return;
if(pend2.pos<=i*1024+j1-1+64)return;
r12_emit();
}
}

评测结果