OSchip
/
OpenBLAS
Not watched
1
0
Fork 0
Code
Releases 66 Wiki evaluate Activity
Issues 0 Pull Requests 0 Datasets Model Cloudbrain HPC
Browse Source

sgemm_kernel_16x4_haswell.S minor changes

tags/v0.2.9.rc1
wernsaar 13 years ago
parent
1cb9579cd0
commit
2b8ab8f55b
1 changed files with 11 additions and 11 deletions
Unified View
Diff Options
Show Stats Download Patch File Download Diff File
  1. +11
    -11
      kernel/x86_64/sgemm_kernel_16x4_haswell.S

+ 11
- 11
kernel/x86_64/sgemm_kernel_16x4_haswell.S View File

@@ -930,7 +930,7 @@
leaq BUFFER1, BO // first buffer to BO leaq BUFFER1, BO // first buffer to BO
addq $4 * SIZE, BO addq $4 * SIZE, BO
movq %rax, BI // Index for BO movq %rax, BI // Index for BO
salq $2 ,BI // BI = BI * 4 ; number of values
leaq (,BI, 4), BI // BI = BI * 4 ; number of values
leaq (BO, BI, SIZE), BO leaq (BO, BI, SIZE), BO
salq $4, %rax // rax = rax * 16 ; number of values salq $4, %rax // rax = rax * 16 ; number of values
leaq (AO, %rax, SIZE), AO leaq (AO, %rax, SIZE), AO
@@ -1053,7 +1053,7 @@
movq K, %rax movq K, %rax
subq KKK, %rax subq KKK, %rax
movq %rax, BI // Index for BO movq %rax, BI // Index for BO
salq $2, BI // BI = BI * 4 ; number of values
leaq (,BI, 4), BI // BI = BI * 4 ; number of values
leaq (BO, BI, SIZE), BO leaq (BO, BI, SIZE), BO
salq $4, %rax // rax = rax * 16 ; number of values salq $4, %rax // rax = rax * 16 ; number of values
leaq (AO, %rax, SIZE), AO leaq (AO, %rax, SIZE), AO
@@ -1096,7 +1096,7 @@
leaq BUFFER1, BO // first buffer to BO leaq BUFFER1, BO // first buffer to BO
addq $4 * SIZE, BO addq $4 * SIZE, BO
movq %rax, BI // Index for BO movq %rax, BI // Index for BO
salq $2, BI // BI = BI * 4 ; number of values
leaq (,BI, 4), BI // BI = BI * 4 ; number of values
leaq (BO, BI, SIZE), BO leaq (BO, BI, SIZE), BO
salq $3, %rax // rax = rax * 8 ; number of values salq $3, %rax // rax = rax * 8 ; number of values
leaq (AO, %rax, SIZE), AO leaq (AO, %rax, SIZE), AO
@@ -1200,7 +1200,7 @@
movq K, %rax movq K, %rax
subq KKK, %rax subq KKK, %rax
movq %rax, BI // Index for BO movq %rax, BI // Index for BO
salq $2, BI // BI = BI * 4 ; number of values
leaq (,BI, 4), BI // BI = BI * 4 ; number of values
leaq (BO, BI, SIZE), BO leaq (BO, BI, SIZE), BO
salq $3, %rax // rax = rax * 8 ; number of values salq $3, %rax // rax = rax * 8 ; number of values
leaq (AO, %rax, SIZE), AO leaq (AO, %rax, SIZE), AO
@@ -1236,7 +1236,7 @@
leaq BUFFER1, BO // first buffer to BO leaq BUFFER1, BO // first buffer to BO
addq $4 * SIZE, BO addq $4 * SIZE, BO
movq %rax, BI // Index for BO movq %rax, BI // Index for BO
salq $2, BI // BI = BI * 4 ; number of values
leaq (,BI, 4), BI // BI = BI * 4 ; number of values
leaq (BO, BI, SIZE), BO leaq (BO, BI, SIZE), BO
salq $2, %rax // rax = rax * 4 ; number of values salq $2, %rax // rax = rax * 4 ; number of values
leaq (AO, %rax, SIZE), AO leaq (AO, %rax, SIZE), AO
@@ -1340,7 +1340,7 @@
movq K, %rax movq K, %rax
subq KKK, %rax subq KKK, %rax
movq %rax, BI // Index for BO movq %rax, BI // Index for BO
salq $2, BI // BI = BI * 4 ; number of values
leaq (,BI, 4), BI // BI = BI * 4 ; number of values
leaq (BO, BI, SIZE), BO leaq (BO, BI, SIZE), BO
salq $2, %rax // rax = rax * 4 ; number of values salq $2, %rax // rax = rax * 4 ; number of values
leaq (AO, %rax, SIZE), AO leaq (AO, %rax, SIZE), AO
@@ -1373,7 +1373,7 @@
leaq BUFFER1, BO // first buffer to BO leaq BUFFER1, BO // first buffer to BO
addq $4 * SIZE, BO addq $4 * SIZE, BO
movq %rax, BI // Index for BO movq %rax, BI // Index for BO
salq $2, BI // BI = BI * 4 ; number of values
leaq (,BI, 4), BI // BI = BI * 4 ; number of values
leaq (BO, BI, SIZE), BO leaq (BO, BI, SIZE), BO
salq $1, %rax // rax = rax * 2 ; number of values salq $1, %rax // rax = rax * 2 ; number of values
leaq (AO, %rax, SIZE), AO leaq (AO, %rax, SIZE), AO
@@ -1451,7 +1451,7 @@
je .L4_39 je .L4_39
movq %rax, BI // Index for BO movq %rax, BI // Index for BO
salq $2, BI // BI = BI * 4 ; number of values
leaq (,BI, 4), BI // BI = BI * 4 ; number of values
salq $1, %rax // rax = rax *2 ; number of values salq $1, %rax // rax = rax *2 ; number of values
leaq (AO, %rax, SIZE), AO leaq (AO, %rax, SIZE), AO
@@ -1477,7 +1477,7 @@
movq K, %rax movq K, %rax
subq KKK, %rax subq KKK, %rax
movq %rax, BI // Index for BO movq %rax, BI // Index for BO
salq $2, BI // BI = BI * 4 ; number of values
leaq (,BI, 4), BI // BI = BI * 4 ; number of values
leaq (BO, BI, SIZE), BO leaq (BO, BI, SIZE), BO
salq $1, %rax // rax = rax * 2 ; number of values salq $1, %rax // rax = rax * 2 ; number of values
leaq (AO, %rax, SIZE), AO leaq (AO, %rax, SIZE), AO
@@ -1509,7 +1509,7 @@
leaq BUFFER1, BO // first buffer to BO leaq BUFFER1, BO // first buffer to BO
addq $4 * SIZE, BO addq $4 * SIZE, BO
movq %rax, BI // Index for BO movq %rax, BI // Index for BO
salq $2, BI // BI = BI * 4 ; number of values
leaq (,BI, 4), BI // BI = BI * 4 ; number of values
leaq (BO, BI, SIZE), BO leaq (BO, BI, SIZE), BO
leaq (AO, %rax, SIZE), AO leaq (AO, %rax, SIZE), AO
#endif #endif
@@ -1609,7 +1609,7 @@
movq K, %rax movq K, %rax
subq KKK, %rax subq KKK, %rax
movq %rax, BI // Index for BO movq %rax, BI // Index for BO
salq $2, BI // BI = BI * 4 ; number of values
leaq (,BI, 4), BI // BI = BI * 4 ; number of values
leaq (BO, BI, SIZE), BO leaq (BO, BI, SIZE), BO
leaq (AO, %rax, SIZE), AO leaq (AO, %rax, SIZE), AO
#endif #endif


Write Preview
Loading…
Cancel
Save