vulkan: implement ADD1, ARANGE, FILL, SOFTPLUS, STEP, ROUND, CEIL, FLOOR, TRUNC (#17319)
* vulkan: initialize array * vulkan: implement ADD1 * vulkan: implement ARANGE * vulkan: implement FILL * vulkan: implement SOFTPLUS * vulkan: implement STEP * vulkan: implement ROUND * vulkan: implement CEIL * vulkan: implement FLOOR * vulkan: implement TRUNC * docs: update Vulkan ops Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
This commit is contained in:
Giuseppe Scrivano
GitHub
parent
1fa4551af0
commit
7d77f07325
18
docs/ops.md
18
docs/ops.md
|
|
@ -17,12 +17,12 @@ Legend:
|
|||
| ABS | ❌ | ✅ | ✅ | 🟡 | 🟡 | ❌ | ✅ | 🟡 | ❌ |
|
||||
| ACC | ❌ | ✅ | ✅ | ✅ | ✅ | ❌ | ✅ | ✅ | ❌ |
|
||||
| ADD | ❌ | ✅ | ✅ | ✅ | 🟡 | 🟡 | ✅ | ✅ | ❌ |
|
||||
| ADD1 | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ | ✅ | ❌ | ❌ |
|
||||
| ADD1 | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ | ✅ | ✅ | ❌ |
|
||||
| ADD_ID | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ | ✅ | ❌ |
|
||||
| ARANGE | ❌ | ✅ | ✅ | ✅ | ✅ | ❌ | ✅ | ❌ | ❌ |
|
||||
| ARANGE | ❌ | ✅ | ✅ | ✅ | ✅ | ❌ | ✅ | ✅ | ❌ |
|
||||
| ARGMAX | ❌ | ✅ | ✅ | ✅ | ✅ | ❌ | ✅ | ✅ | ❌ |
|
||||
| ARGSORT | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | 🟡 | ❌ |
|
||||
| CEIL | ❌ | ❌ | ✅ | 🟡 | ❌ | ❌ | 🟡 | ❌ | ❌ |
|
||||
| CEIL | ❌ | ❌ | ✅ | 🟡 | ❌ | ❌ | 🟡 | 🟡 | ❌ |
|
||||
| CLAMP | ❌ | ✅ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | 🟡 | ❌ |
|
||||
| CONCAT | ❌ | ✅ | ✅ | 🟡 | ✅ | 🟡 | ✅ | ✅ | ❌ |
|
||||
| CONT | ❌ | 🟡 | ✅ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | ❌ |
|
||||
|
|
@ -43,9 +43,9 @@ Legend:
|
|||
| ELU | ❌ | ✅ | ✅ | 🟡 | 🟡 | ❌ | ✅ | ❌ | ❌ |
|
||||
| EXP | ❌ | ✅ | ✅ | 🟡 | 🟡 | ❌ | ✅ | 🟡 | ❌ |
|
||||
| EXPM1 | ❌ | ❌ | ✅ | 🟡 | ❌ | ❌ | ❌ | ❌ | ❌ |
|
||||
| FILL | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
|
||||
| FILL | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ |
|
||||
| FLASH_ATTN_EXT | ❌ | 🟡 | ✅ | 🟡 | 🟡 | ❌ | ❌ | 🟡 | ❌ |
|
||||
| FLOOR | ❌ | ❌ | ✅ | 🟡 | ❌ | ❌ | 🟡 | ❌ | ❌ |
|
||||
| FLOOR | ❌ | ❌ | ✅ | 🟡 | ❌ | ❌ | 🟡 | 🟡 | ❌ |
|
||||
| GATED_LINEAR_ATTN | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ✅ | ❌ | ❌ |
|
||||
| GEGLU | ❌ | ✅ | ✅ | ✅ | 🟡 | ✅ | ✅ | 🟡 | ❌ |
|
||||
| GEGLU_ERF | ❌ | ✅ | ✅ | ✅ | 🟡 | ✅ | ✅ | 🟡 | ❌ |
|
||||
|
|
@ -87,7 +87,7 @@ Legend:
|
|||
| ROLL | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ✅ | ✅ | ❌ |
|
||||
| ROPE | ❌ | 🟡 | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ |
|
||||
| ROPE_BACK | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ | ✅ | ❌ |
|
||||
| ROUND | ❌ | ❌ | ✅ | 🟡 | ❌ | ❌ | 🟡 | ❌ | ❌ |
|
||||
| ROUND | ❌ | ❌ | ✅ | 🟡 | ❌ | ❌ | 🟡 | 🟡 | ❌ |
|
||||
| RWKV_WKV6 | ❌ | ❌ | ✅ | ✅ | ✅ | ❌ | ✅ | ✅ | ❌ |
|
||||
| RWKV_WKV7 | ❌ | ❌ | ✅ | ✅ | ✅ | ❌ | ✅ | ✅ | ❌ |
|
||||
| SCALE | ❌ | 🟡 | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ |
|
||||
|
|
@ -99,7 +99,7 @@ Legend:
|
|||
| SILU_BACK | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ | ✅ | ❌ |
|
||||
| SIN | ❌ | ✅ | ✅ | ✅ | 🟡 | ❌ | 🟡 | 🟡 | ❌ |
|
||||
| SOFTCAP | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
|
||||
| SOFTPLUS | ❌ | ❌ | ✅ | 🟡 | ❌ | ❌ | ❌ | ❌ | ❌ |
|
||||
| SOFTPLUS | ❌ | ❌ | ✅ | 🟡 | ❌ | ❌ | ❌ | 🟡 | ❌ |
|
||||
| SOFT_MAX | ❌ | 🟡 | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ |
|
||||
| SOFT_MAX_BACK | ❌ | ❌ | 🟡 | 🟡 | ❌ | ❌ | 🟡 | ✅ | ❌ |
|
||||
| SOLVE_TRI | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
|
||||
|
|
@ -107,7 +107,7 @@ Legend:
|
|||
| SQRT | ❌ | ✅ | ✅ | ✅ | 🟡 | ❌ | 🟡 | 🟡 | ❌ |
|
||||
| SSM_CONV | ❌ | ❌ | ✅ | ✅ | ✅ | ❌ | ✅ | ✅ | ❌ |
|
||||
| SSM_SCAN | ❌ | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ | 🟡 | ❌ |
|
||||
| STEP | ❌ | ✅ | ✅ | 🟡 | 🟡 | ❌ | ✅ | ❌ | ❌ |
|
||||
| STEP | ❌ | ✅ | ✅ | 🟡 | 🟡 | ❌ | ✅ | 🟡 | ❌ |
|
||||
| SUB | ❌ | ✅ | ✅ | ✅ | 🟡 | 🟡 | ✅ | ✅ | ❌ |
|
||||
| SUM | ❌ | ✅ | ✅ | 🟡 | ❌ | ❌ | 🟡 | 🟡 | ❌ |
|
||||
| SUM_ROWS | ❌ | ✅ | ✅ | 🟡 | ✅ | ✅ | 🟡 | ✅ | ❌ |
|
||||
|
|
@ -116,6 +116,6 @@ Legend:
|
|||
| TANH | ❌ | ✅ | ✅ | 🟡 | 🟡 | ✅ | ✅ | 🟡 | ❌ |
|
||||
| TIMESTEP_EMBEDDING | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ |
|
||||
| TRI | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
|
||||
| TRUNC | ❌ | ❌ | ✅ | 🟡 | ❌ | ❌ | 🟡 | ❌ | ❌ |
|
||||
| TRUNC | ❌ | ❌ | ✅ | 🟡 | ❌ | ❌ | 🟡 | 🟡 | ❌ |
|
||||
| UPSCALE | ❌ | 🟡 | ✅ | ✅ | 🟡 | ✅ | 🟡 | ✅ | ❌ |
|
||||
| XIELU | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
|
||||
|
|
|
|||
|
|
@ -5,8 +5,8 @@
|
|||
"Vulkan0","SGN","type=f16,ne_a=[5,7,11,13],v=0","support","0","no","Vulkan"
|
||||
"Vulkan0","NEG","type=f16,ne_a=[128,2,2,2],v=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","NEG","type=f16,ne_a=[5,7,11,13],v=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","STEP","type=f16,ne_a=[128,2,2,2],v=0","support","0","no","Vulkan"
|
||||
"Vulkan0","STEP","type=f16,ne_a=[5,7,11,13],v=0","support","0","no","Vulkan"
|
||||
"Vulkan0","STEP","type=f16,ne_a=[128,2,2,2],v=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","STEP","type=f16,ne_a=[5,7,11,13],v=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","TANH","type=f16,ne_a=[128,2,2,2],v=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","TANH","type=f16,ne_a=[5,7,11,13],v=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","ELU","type=f16,ne_a=[128,2,2,2],v=0","support","0","no","Vulkan"
|
||||
|
|
@ -29,18 +29,18 @@
|
|||
"Vulkan0","EXP","type=f16,ne_a=[5,7,11,13],v=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","EXPM1","type=f16,ne_a=[128,2,2,2],v=0","support","0","no","Vulkan"
|
||||
"Vulkan0","EXPM1","type=f16,ne_a=[5,7,11,13],v=0","support","0","no","Vulkan"
|
||||
"Vulkan0","SOFTPLUS","type=f16,ne_a=[128,2,2,2],v=0","support","0","no","Vulkan"
|
||||
"Vulkan0","SOFTPLUS","type=f16,ne_a=[5,7,11,13],v=0","support","0","no","Vulkan"
|
||||
"Vulkan0","SOFTPLUS","type=f16,ne_a=[128,2,2,2],v=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","SOFTPLUS","type=f16,ne_a=[5,7,11,13],v=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","GELU_ERF","type=f16,ne_a=[128,2,2,2],v=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","GELU_ERF","type=f16,ne_a=[5,7,11,13],v=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","FLOOR","type=f16,ne_a=[128,2,2,2],v=0","support","0","no","Vulkan"
|
||||
"Vulkan0","FLOOR","type=f16,ne_a=[5,7,11,13],v=0","support","0","no","Vulkan"
|
||||
"Vulkan0","CEIL","type=f16,ne_a=[128,2,2,2],v=0","support","0","no","Vulkan"
|
||||
"Vulkan0","CEIL","type=f16,ne_a=[5,7,11,13],v=0","support","0","no","Vulkan"
|
||||
"Vulkan0","ROUND","type=f16,ne_a=[128,2,2,2],v=0","support","0","no","Vulkan"
|
||||
"Vulkan0","ROUND","type=f16,ne_a=[5,7,11,13],v=0","support","0","no","Vulkan"
|
||||
"Vulkan0","TRUNC","type=f16,ne_a=[128,2,2,2],v=0","support","0","no","Vulkan"
|
||||
"Vulkan0","TRUNC","type=f16,ne_a=[5,7,11,13],v=0","support","0","no","Vulkan"
|
||||
"Vulkan0","FLOOR","type=f16,ne_a=[128,2,2,2],v=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","FLOOR","type=f16,ne_a=[5,7,11,13],v=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","CEIL","type=f16,ne_a=[128,2,2,2],v=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","CEIL","type=f16,ne_a=[5,7,11,13],v=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","ROUND","type=f16,ne_a=[128,2,2,2],v=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","ROUND","type=f16,ne_a=[5,7,11,13],v=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","TRUNC","type=f16,ne_a=[128,2,2,2],v=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","TRUNC","type=f16,ne_a=[5,7,11,13],v=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","ABS","type=f16,ne_a=[128,2,2,2],v=1","support","0","no","Vulkan"
|
||||
"Vulkan0","ABS","type=f16,ne_a=[5,7,11,13],v=1","support","0","no","Vulkan"
|
||||
"Vulkan0","SGN","type=f16,ne_a=[128,2,2,2],v=1","support","0","no","Vulkan"
|
||||
|
|
@ -89,8 +89,8 @@
|
|||
"Vulkan0","SGN","type=f32,ne_a=[5,7,11,13],v=0","support","0","no","Vulkan"
|
||||
"Vulkan0","NEG","type=f32,ne_a=[128,2,2,2],v=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","NEG","type=f32,ne_a=[5,7,11,13],v=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","STEP","type=f32,ne_a=[128,2,2,2],v=0","support","0","no","Vulkan"
|
||||
"Vulkan0","STEP","type=f32,ne_a=[5,7,11,13],v=0","support","0","no","Vulkan"
|
||||
"Vulkan0","STEP","type=f32,ne_a=[128,2,2,2],v=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","STEP","type=f32,ne_a=[5,7,11,13],v=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","TANH","type=f32,ne_a=[128,2,2,2],v=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","TANH","type=f32,ne_a=[5,7,11,13],v=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","ELU","type=f32,ne_a=[128,2,2,2],v=0","support","0","no","Vulkan"
|
||||
|
|
@ -113,18 +113,18 @@
|
|||
"Vulkan0","EXP","type=f32,ne_a=[5,7,11,13],v=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","EXPM1","type=f32,ne_a=[128,2,2,2],v=0","support","0","no","Vulkan"
|
||||
"Vulkan0","EXPM1","type=f32,ne_a=[5,7,11,13],v=0","support","0","no","Vulkan"
|
||||
"Vulkan0","SOFTPLUS","type=f32,ne_a=[128,2,2,2],v=0","support","0","no","Vulkan"
|
||||
"Vulkan0","SOFTPLUS","type=f32,ne_a=[5,7,11,13],v=0","support","0","no","Vulkan"
|
||||
"Vulkan0","SOFTPLUS","type=f32,ne_a=[128,2,2,2],v=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","SOFTPLUS","type=f32,ne_a=[5,7,11,13],v=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","GELU_ERF","type=f32,ne_a=[128,2,2,2],v=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","GELU_ERF","type=f32,ne_a=[5,7,11,13],v=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","FLOOR","type=f32,ne_a=[128,2,2,2],v=0","support","0","no","Vulkan"
|
||||
"Vulkan0","FLOOR","type=f32,ne_a=[5,7,11,13],v=0","support","0","no","Vulkan"
|
||||
"Vulkan0","CEIL","type=f32,ne_a=[128,2,2,2],v=0","support","0","no","Vulkan"
|
||||
"Vulkan0","CEIL","type=f32,ne_a=[5,7,11,13],v=0","support","0","no","Vulkan"
|
||||
"Vulkan0","ROUND","type=f32,ne_a=[128,2,2,2],v=0","support","0","no","Vulkan"
|
||||
"Vulkan0","ROUND","type=f32,ne_a=[5,7,11,13],v=0","support","0","no","Vulkan"
|
||||
"Vulkan0","TRUNC","type=f32,ne_a=[128,2,2,2],v=0","support","0","no","Vulkan"
|
||||
"Vulkan0","TRUNC","type=f32,ne_a=[5,7,11,13],v=0","support","0","no","Vulkan"
|
||||
"Vulkan0","FLOOR","type=f32,ne_a=[128,2,2,2],v=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","FLOOR","type=f32,ne_a=[5,7,11,13],v=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","CEIL","type=f32,ne_a=[128,2,2,2],v=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","CEIL","type=f32,ne_a=[5,7,11,13],v=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","ROUND","type=f32,ne_a=[128,2,2,2],v=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","ROUND","type=f32,ne_a=[5,7,11,13],v=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","TRUNC","type=f32,ne_a=[128,2,2,2],v=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","TRUNC","type=f32,ne_a=[5,7,11,13],v=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","ABS","type=f32,ne_a=[128,2,2,2],v=1","support","0","no","Vulkan"
|
||||
"Vulkan0","ABS","type=f32,ne_a=[5,7,11,13],v=1","support","0","no","Vulkan"
|
||||
"Vulkan0","SGN","type=f32,ne_a=[128,2,2,2],v=1","support","0","no","Vulkan"
|
||||
|
|
@ -5654,7 +5654,7 @@
|
|||
"Vulkan0","SUB","type=f32,ne=[64,262144,1,1],nr=[1,1,1,1],nf=1","support","1","yes","Vulkan"
|
||||
"Vulkan0","MUL","type=f32,ne=[64,262144,1,1],nr=[1,1,1,1],nf=1","support","1","yes","Vulkan"
|
||||
"Vulkan0","DIV","type=f32,ne=[64,262144,1,1],nr=[1,1,1,1],nf=1","support","1","yes","Vulkan"
|
||||
"Vulkan0","ADD1","type=f32,ne=[10,5,4,3]","support","0","no","Vulkan"
|
||||
"Vulkan0","ADD1","type=f32,ne=[10,5,4,3]","support","1","yes","Vulkan"
|
||||
"Vulkan0","SCALE","type=f32,ne=[10,10,10,10],scale=2.000000,bias=0.000000,inplace=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","SCALE","type=f32,ne=[10,10,10,10],scale=2.000000,bias=1.000000,inplace=0","support","1","yes","Vulkan"
|
||||
"Vulkan0","SCALE","type=f32,ne=[10,10,10,10],scale=2.000000,bias=1.000000,inplace=1","support","1","yes","Vulkan"
|
||||
|
|
@ -8632,10 +8632,10 @@
|
|||
"Vulkan0","COS","type=f16,ne=[10,2,2,2]","support","0","no","Vulkan"
|
||||
"Vulkan0","CLAMP","type=f16,ne=[10,5,4,3],min=-0.500000,max=0.500000","support","0","no","Vulkan"
|
||||
"Vulkan0","LEAKY_RELU","type=f16,ne_a=[10,5,4,3],negative_slope=0.100000","support","0","no","Vulkan"
|
||||
"Vulkan0","FLOOR","type=f16,ne=[10,2,2,2]","support","0","no","Vulkan"
|
||||
"Vulkan0","CEIL","type=f16,ne=[10,2,2,2]","support","0","no","Vulkan"
|
||||
"Vulkan0","ROUND","type=f16,ne=[10,2,2,2]","support","0","no","Vulkan"
|
||||
"Vulkan0","TRUNC","type=f16,ne=[10,2,2,2]","support","0","no","Vulkan"
|
||||
"Vulkan0","FLOOR","type=f16,ne=[10,2,2,2]","support","1","yes","Vulkan"
|
||||
"Vulkan0","CEIL","type=f16,ne=[10,2,2,2]","support","1","yes","Vulkan"
|
||||
"Vulkan0","ROUND","type=f16,ne=[10,2,2,2]","support","1","yes","Vulkan"
|
||||
"Vulkan0","TRUNC","type=f16,ne=[10,2,2,2]","support","1","yes","Vulkan"
|
||||
"Vulkan0","SQR","type=f16,ne=[7,1,5,3]","support","0","no","Vulkan"
|
||||
"Vulkan0","SQRT","type=f16,ne=[7,1,5,3]","support","0","no","Vulkan"
|
||||
"Vulkan0","LOG","type=f16,ne=[7,1,5,3]","support","1","yes","Vulkan"
|
||||
|
|
@ -8643,10 +8643,10 @@
|
|||
"Vulkan0","COS","type=f16,ne=[7,1,5,3]","support","0","no","Vulkan"
|
||||
"Vulkan0","CLAMP","type=f16,ne=[7,1,5,3],min=-0.500000,max=0.500000","support","0","no","Vulkan"
|
||||
"Vulkan0","LEAKY_RELU","type=f16,ne_a=[7,1,5,3],negative_slope=0.100000","support","0","no","Vulkan"
|
||||
"Vulkan0","FLOOR","type=f16,ne=[7,1,5,3]","support","0","no","Vulkan"
|
||||
"Vulkan0","CEIL","type=f16,ne=[7,1,5,3]","support","0","no","Vulkan"
|
||||
"Vulkan0","ROUND","type=f16,ne=[7,1,5,3]","support","0","no","Vulkan"
|
||||
"Vulkan0","TRUNC","type=f16,ne=[7,1,5,3]","support","0","no","Vulkan"
|
||||
"Vulkan0","FLOOR","type=f16,ne=[7,1,5,3]","support","1","yes","Vulkan"
|
||||
"Vulkan0","CEIL","type=f16,ne=[7,1,5,3]","support","1","yes","Vulkan"
|
||||
"Vulkan0","ROUND","type=f16,ne=[7,1,5,3]","support","1","yes","Vulkan"
|
||||
"Vulkan0","TRUNC","type=f16,ne=[7,1,5,3]","support","1","yes","Vulkan"
|
||||
"Vulkan0","SQR","type=f32,ne=[10,5,4,3]","support","1","yes","Vulkan"
|
||||
"Vulkan0","SQRT","type=f32,ne=[10,3,3,2]","support","1","yes","Vulkan"
|
||||
"Vulkan0","LOG","type=f32,ne=[10,5,4,3]","support","1","yes","Vulkan"
|
||||
|
|
@ -8654,10 +8654,10 @@
|
|||
"Vulkan0","COS","type=f32,ne=[10,2,2,2]","support","1","yes","Vulkan"
|
||||
"Vulkan0","CLAMP","type=f32,ne=[10,5,4,3],min=-0.500000,max=0.500000","support","1","yes","Vulkan"
|
||||
"Vulkan0","LEAKY_RELU","type=f32,ne_a=[10,5,4,3],negative_slope=0.100000","support","1","yes","Vulkan"
|
||||
"Vulkan0","FLOOR","type=f32,ne=[10,2,2,2]","support","0","no","Vulkan"
|
||||
"Vulkan0","CEIL","type=f32,ne=[10,2,2,2]","support","0","no","Vulkan"
|
||||
"Vulkan0","ROUND","type=f32,ne=[10,2,2,2]","support","0","no","Vulkan"
|
||||
"Vulkan0","TRUNC","type=f32,ne=[10,2,2,2]","support","0","no","Vulkan"
|
||||
"Vulkan0","FLOOR","type=f32,ne=[10,2,2,2]","support","1","yes","Vulkan"
|
||||
"Vulkan0","CEIL","type=f32,ne=[10,2,2,2]","support","1","yes","Vulkan"
|
||||
"Vulkan0","ROUND","type=f32,ne=[10,2,2,2]","support","1","yes","Vulkan"
|
||||
"Vulkan0","TRUNC","type=f32,ne=[10,2,2,2]","support","1","yes","Vulkan"
|
||||
"Vulkan0","SQR","type=f32,ne=[7,1,5,3]","support","1","yes","Vulkan"
|
||||
"Vulkan0","SQRT","type=f32,ne=[7,1,5,3]","support","1","yes","Vulkan"
|
||||
"Vulkan0","LOG","type=f32,ne=[7,1,5,3]","support","1","yes","Vulkan"
|
||||
|
|
@ -8665,10 +8665,10 @@
|
|||
"Vulkan0","COS","type=f32,ne=[7,1,5,3]","support","1","yes","Vulkan"
|
||||
"Vulkan0","CLAMP","type=f32,ne=[7,1,5,3],min=-0.500000,max=0.500000","support","1","yes","Vulkan"
|
||||
"Vulkan0","LEAKY_RELU","type=f32,ne_a=[7,1,5,3],negative_slope=0.100000","support","1","yes","Vulkan"
|
||||
"Vulkan0","FLOOR","type=f32,ne=[7,1,5,3]","support","0","no","Vulkan"
|
||||
"Vulkan0","CEIL","type=f32,ne=[7,1,5,3]","support","0","no","Vulkan"
|
||||
"Vulkan0","ROUND","type=f32,ne=[7,1,5,3]","support","0","no","Vulkan"
|
||||
"Vulkan0","TRUNC","type=f32,ne=[7,1,5,3]","support","0","no","Vulkan"
|
||||
"Vulkan0","FLOOR","type=f32,ne=[7,1,5,3]","support","1","yes","Vulkan"
|
||||
"Vulkan0","CEIL","type=f32,ne=[7,1,5,3]","support","1","yes","Vulkan"
|
||||
"Vulkan0","ROUND","type=f32,ne=[7,1,5,3]","support","1","yes","Vulkan"
|
||||
"Vulkan0","TRUNC","type=f32,ne=[7,1,5,3]","support","1","yes","Vulkan"
|
||||
"Vulkan0","DIAG_MASK_INF","type=f32,ne=[10,10,1,1],n_past=5","support","1","yes","Vulkan"
|
||||
"Vulkan0","DIAG_MASK_INF","type=f32,ne=[10,10,3,1],n_past=5","support","1","yes","Vulkan"
|
||||
"Vulkan0","DIAG_MASK_INF","type=f32,ne=[10,10,3,2],n_past=5","support","1","yes","Vulkan"
|
||||
|
|
@ -9478,7 +9478,7 @@
|
|||
"Vulkan0","PAD_REFLECT_1D","type=f32,ne_a=[512,34,2,1],pad_0=10,pad_1=9","support","0","no","Vulkan"
|
||||
"Vulkan0","PAD_REFLECT_1D","type=f32,ne_a=[3000,384,4,1],pad_0=10,pad_1=9","support","0","no","Vulkan"
|
||||
"Vulkan0","ROLL","shift0=3,shift1=-2,shift3=1,shift4=-1","support","1","yes","Vulkan"
|
||||
"Vulkan0","ARANGE","type=f32,start=0.000000,stop=10.000000,step=1.000000","support","0","no","Vulkan"
|
||||
"Vulkan0","ARANGE","type=f32,start=0.000000,stop=10.000000,step=1.000000","support","1","yes","Vulkan"
|
||||
"Vulkan0","TIMESTEP_EMBEDDING","type=f32,ne_a=[2,1,1,1],dim=320,max_period=10000","support","1","yes","Vulkan"
|
||||
"Vulkan0","LEAKY_RELU","type=f32,ne_a=[10,5,4,3],negative_slope=0.100000","support","1","yes","Vulkan"
|
||||
"Vulkan0","CUMSUM","type=f32,ne=[10,5,4,3]","support","0","no","Vulkan"
|
||||
|
|
@ -9487,9 +9487,9 @@
|
|||
"Vulkan0","TRI","type=f32,ne=[10,10,4,3],tri_type=2","support","0","no","Vulkan"
|
||||
"Vulkan0","TRI","type=f32,ne=[10,10,4,3],tri_type=1","support","0","no","Vulkan"
|
||||
"Vulkan0","TRI","type=f32,ne=[10,10,4,3],tri_type=0","support","0","no","Vulkan"
|
||||
"Vulkan0","FILL","type=f32,ne=[10,10,4,3],c=0.000000","support","0","no","Vulkan"
|
||||
"Vulkan0","FILL","type=f32,ne=[303,207,11,3],c=2.000000","support","0","no","Vulkan"
|
||||
"Vulkan0","FILL","type=f32,ne=[800,600,4,4],c=-152.000000","support","0","no","Vulkan"
|
||||
"Vulkan0","FILL","type=f32,ne=[10,10,4,3],c=0.000000","support","1","yes","Vulkan"
|
||||
"Vulkan0","FILL","type=f32,ne=[303,207,11,3],c=2.000000","support","1","yes","Vulkan"
|
||||
"Vulkan0","FILL","type=f32,ne=[800,600,4,4],c=-152.000000","support","1","yes","Vulkan"
|
||||
"Vulkan0","SOLVE_TRI","type=f32,ne_lhs=[10,10,4,3],ne_rhs=[3,10,4,3]","support","0","no","Vulkan"
|
||||
"Vulkan0","SOLVE_TRI","type=f32,ne_lhs=[11,11,1,1],ne_rhs=[5,11,1,1]","support","0","no","Vulkan"
|
||||
"Vulkan0","SOLVE_TRI","type=f32,ne_lhs=[17,17,2,4],ne_rhs=[9,17,2,4]","support","0","no","Vulkan"
|
||||
|
|
|
|||
|
Can't render this file because it is too large.
|
|
|
@ -669,6 +669,20 @@ struct vk_device_struct {
|
|||
vk_pipeline pipeline_hardsigmoid[2];
|
||||
vk_pipeline pipeline_hardswish[2];
|
||||
vk_pipeline pipeline_abs[2];
|
||||
vk_pipeline pipeline_softplus[2];
|
||||
vk_pipeline pipeline_step[2];
|
||||
vk_pipeline pipeline_round[2];
|
||||
vk_pipeline pipeline_ceil[2];
|
||||
vk_pipeline pipeline_floor[2];
|
||||
vk_pipeline pipeline_trunc[2];
|
||||
|
||||
vk_pipeline pipeline_add1_f16_f16;
|
||||
vk_pipeline pipeline_add1_f16_f32;
|
||||
vk_pipeline pipeline_add1_f32_f32;
|
||||
|
||||
vk_pipeline pipeline_arange_f32;
|
||||
|
||||
vk_pipeline pipeline_fill_f32;
|
||||
|
||||
vk_pipeline pipeline_geglu[2];
|
||||
vk_pipeline pipeline_reglu[2];
|
||||
|
|
@ -3841,6 +3855,12 @@ static void ggml_vk_load_shaders(vk_device& device) {
|
|||
CREATE_UNARY(hardsigmoid)
|
||||
CREATE_UNARY(hardswish)
|
||||
CREATE_UNARY(abs)
|
||||
CREATE_UNARY(softplus)
|
||||
CREATE_UNARY(step)
|
||||
CREATE_UNARY(round)
|
||||
CREATE_UNARY(ceil)
|
||||
CREATE_UNARY(floor)
|
||||
CREATE_UNARY(trunc)
|
||||
#undef CREATE_UNARY
|
||||
|
||||
#define CREATE_UNARY_RTE(name) \
|
||||
|
|
@ -3854,6 +3874,14 @@ static void ggml_vk_load_shaders(vk_device& device) {
|
|||
CREATE_UNARY_RTE(exp)
|
||||
#undef CREATE_UNARY_RTE
|
||||
|
||||
ggml_vk_create_pipeline(device, device->pipeline_add1_f16_f16, "add1_f16_f16", add1_f16_f16_len, add1_f16_f16_data, "main", 3, sizeof(vk_op_binary_push_constants), {512, 1, 1}, {}, 1);
|
||||
ggml_vk_create_pipeline(device, device->pipeline_add1_f16_f32, "add1_f16_f32", add1_f16_f32_len, add1_f16_f32_data, "main", 3, sizeof(vk_op_binary_push_constants), {512, 1, 1}, {}, 1);
|
||||
ggml_vk_create_pipeline(device, device->pipeline_add1_f32_f32, "add1_f32_f32", add1_f32_f32_len, add1_f32_f32_data, "main", 3, sizeof(vk_op_binary_push_constants), {512, 1, 1}, {}, 1);
|
||||
|
||||
ggml_vk_create_pipeline(device, device->pipeline_arange_f32, "arange_f32", arange_f32_len, arange_f32_data, "main", 1, sizeof(vk_op_unary_push_constants), {512, 1, 1}, {}, 1);
|
||||
|
||||
ggml_vk_create_pipeline(device, device->pipeline_fill_f32, "fill_f32", fill_f32_len, fill_f32_data, "main", 1, sizeof(vk_op_unary_push_constants), {512, 1, 1}, {}, 1);
|
||||
|
||||
#define CREATE_GLU(name) \
|
||||
if (device->float_controls_rte_fp16) { \
|
||||
ggml_vk_create_pipeline(device, device->pipeline_ ## name [0], #name "_f32_rte", name ## _f32_rte_len, name ## _f32_rte_data, "main", 3, sizeof(vk_op_glu_push_constants), {512, 1, 1}, {}, 1, true); \
|
||||
|
|
@ -8279,6 +8307,18 @@ static vk_pipeline ggml_vk_op_get_pipeline(ggml_backend_vk_context * ctx, const
|
|||
return ctx->device->pipeline_hardswish[dst->type == GGML_TYPE_F16];
|
||||
case GGML_UNARY_OP_ABS:
|
||||
return ctx->device->pipeline_abs[dst->type == GGML_TYPE_F16];
|
||||
case GGML_UNARY_OP_SOFTPLUS:
|
||||
return ctx->device->pipeline_softplus[dst->type == GGML_TYPE_F16];
|
||||
case GGML_UNARY_OP_STEP:
|
||||
return ctx->device->pipeline_step[dst->type == GGML_TYPE_F16];
|
||||
case GGML_UNARY_OP_ROUND:
|
||||
return ctx->device->pipeline_round[dst->type == GGML_TYPE_F16];
|
||||
case GGML_UNARY_OP_CEIL:
|
||||
return ctx->device->pipeline_ceil[dst->type == GGML_TYPE_F16];
|
||||
case GGML_UNARY_OP_FLOOR:
|
||||
return ctx->device->pipeline_floor[dst->type == GGML_TYPE_F16];
|
||||
case GGML_UNARY_OP_TRUNC:
|
||||
return ctx->device->pipeline_trunc[dst->type == GGML_TYPE_F16];
|
||||
default:
|
||||
break;
|
||||
}
|
||||
|
|
@ -8473,7 +8513,7 @@ static vk_pipeline ggml_vk_op_get_pipeline(ggml_backend_vk_context * ctx, const
|
|||
case GGML_OP_CONV_TRANSPOSE_2D:
|
||||
if (src1->type == GGML_TYPE_F32 && dst->type == GGML_TYPE_F32 &&
|
||||
ggml_is_contiguous(src0) && ggml_is_contiguous(src1) && ggml_is_contiguous(dst)) {
|
||||
std::array<uint32_t, 3> elements;
|
||||
std::array<uint32_t, 3> elements{};
|
||||
if (op == GGML_OP_CONV_2D) elements = ggml_vk_get_conv_elements(dst);
|
||||
else if (op == GGML_OP_CONV_TRANSPOSE_2D) elements = ggml_vk_get_conv_transpose_2d_elements(dst);
|
||||
vk_conv_shapes shape;
|
||||
|
|
@ -8551,6 +8591,27 @@ static vk_pipeline ggml_vk_op_get_pipeline(ggml_backend_vk_context * ctx, const
|
|||
}
|
||||
}
|
||||
return nullptr;
|
||||
case GGML_OP_ADD1:
|
||||
if (src0->type == GGML_TYPE_F16 && src1->type == GGML_TYPE_F16 && dst->type == GGML_TYPE_F16) {
|
||||
return ctx->device->pipeline_add1_f16_f16;
|
||||
}
|
||||
if (src0->type == GGML_TYPE_F16 && src1->type == GGML_TYPE_F32 && dst->type == GGML_TYPE_F16) {
|
||||
return ctx->device->pipeline_add1_f16_f32;
|
||||
}
|
||||
if (src0->type == GGML_TYPE_F32 && src1->type == GGML_TYPE_F32 && dst->type == GGML_TYPE_F32) {
|
||||
return ctx->device->pipeline_add1_f32_f32;
|
||||
}
|
||||
return nullptr;
|
||||
case GGML_OP_ARANGE:
|
||||
if (dst->type == GGML_TYPE_F32) {
|
||||
return ctx->device->pipeline_arange_f32;
|
||||
}
|
||||
return nullptr;
|
||||
case GGML_OP_FILL:
|
||||
if (dst->type == GGML_TYPE_F32) {
|
||||
return ctx->device->pipeline_fill_f32;
|
||||
}
|
||||
return nullptr;
|
||||
default:
|
||||
return nullptr;
|
||||
}
|
||||
|
|
@ -8840,6 +8901,9 @@ static void ggml_vk_op_f32(ggml_backend_vk_context * ctx, vk_context& subctx, co
|
|||
case GGML_OP_SUB:
|
||||
case GGML_OP_DIV:
|
||||
case GGML_OP_MUL:
|
||||
case GGML_OP_ADD1:
|
||||
case GGML_OP_ARANGE:
|
||||
case GGML_OP_FILL:
|
||||
case GGML_OP_SCALE:
|
||||
case GGML_OP_SQR:
|
||||
case GGML_OP_SQRT:
|
||||
|
|
@ -9457,6 +9521,63 @@ static void ggml_vk_sqrt(ggml_backend_vk_context * ctx, vk_context& subctx, cons
|
|||
ggml_vk_op_f32(ctx, subctx, src0, nullptr, nullptr, nullptr, dst, GGML_OP_SQRT, vk_op_unary_push_constants_init(src0, dst));
|
||||
}
|
||||
|
||||
static void ggml_vk_add1(ggml_backend_vk_context * ctx, vk_context& subctx, const ggml_tensor * src0, const ggml_tensor * src1, ggml_tensor * dst) {
|
||||
const uint32_t src0_type_size = ggml_type_size(src0->type);
|
||||
const uint32_t src1_type_size = ggml_type_size(src1->type);
|
||||
const uint32_t dst_type_size = ggml_type_size(dst->type);
|
||||
|
||||
ggml_vk_op_f32<vk_op_binary_push_constants>(ctx, subctx, src0, src1, nullptr, nullptr, dst, GGML_OP_ADD1, {
|
||||
(uint32_t)ggml_nelements(src0),
|
||||
(uint32_t)src0->ne[0], (uint32_t)src0->ne[1], (uint32_t)src0->ne[2],(uint32_t)src0->ne[3], (uint32_t)src0->nb[0] / src0_type_size, (uint32_t)src0->nb[1] / src0_type_size, (uint32_t)src0->nb[2] / src0_type_size, (uint32_t)src0->nb[3] / src0_type_size,
|
||||
(uint32_t)src1->ne[0], (uint32_t)src1->ne[1], (uint32_t)src1->ne[2],(uint32_t)src1->ne[3], (uint32_t)src1->nb[0] / src1_type_size, (uint32_t)src1->nb[1] / src1_type_size, (uint32_t)src1->nb[2] / src1_type_size, (uint32_t)src1->nb[3] / src1_type_size,
|
||||
(uint32_t) dst->ne[0], (uint32_t) dst->ne[1], (uint32_t) dst->ne[2],(uint32_t) dst->ne[3], (uint32_t) dst->nb[0] / dst_type_size, (uint32_t) dst->nb[1] / dst_type_size, (uint32_t) dst->nb[2] / dst_type_size, (uint32_t) dst->nb[3] / dst_type_size,
|
||||
0,
|
||||
0.0f, 0.0f, 0,
|
||||
});
|
||||
}
|
||||
|
||||
static void ggml_vk_arange(ggml_backend_vk_context * ctx, vk_context& subctx, ggml_tensor * dst) {
|
||||
VK_LOG_DEBUG("ggml_vk_arange(dst=" << dst << ", ne=" << ggml_nelements(dst) << ")");
|
||||
|
||||
vk_op_push_constants pc = {
|
||||
(uint32_t)ggml_nelements(dst),
|
||||
1,
|
||||
ggml_get_op_params_f32(dst, 0),
|
||||
ggml_get_op_params_f32(dst, 2),
|
||||
};
|
||||
|
||||
vk_pipeline pipeline = ggml_vk_op_get_pipeline(ctx, nullptr, nullptr, nullptr, dst, GGML_OP_ARANGE);
|
||||
GGML_ASSERT(pipeline != nullptr);
|
||||
|
||||
ggml_pipeline_request_descriptor_sets(ctx, pipeline, 1);
|
||||
vk_subbuffer dst_buf = ggml_vk_tensor_subbuffer(ctx, dst, false);
|
||||
|
||||
std::array<uint32_t, 3> elements = { (uint32_t)ggml_nelements(dst), 1, 1 };
|
||||
|
||||
ggml_vk_dispatch_pipeline(ctx, subctx, pipeline, { dst_buf }, pc, elements);
|
||||
}
|
||||
|
||||
static void ggml_vk_fill(ggml_backend_vk_context * ctx, vk_context& subctx, ggml_tensor * dst) {
|
||||
VK_LOG_DEBUG("ggml_vk_fill(dst=" << dst << ", ne=" << ggml_nelements(dst) << ")");
|
||||
|
||||
vk_op_push_constants pc = {
|
||||
(uint32_t)ggml_nelements(dst),
|
||||
1,
|
||||
ggml_get_op_params_f32(dst, 0),
|
||||
0.0f,
|
||||
};
|
||||
|
||||
vk_pipeline pipeline = ggml_vk_op_get_pipeline(ctx, nullptr, nullptr, nullptr, dst, GGML_OP_FILL);
|
||||
GGML_ASSERT(pipeline != nullptr);
|
||||
|
||||
ggml_pipeline_request_descriptor_sets(ctx, pipeline, 1);
|
||||
vk_subbuffer dst_buf = ggml_vk_tensor_subbuffer(ctx, dst, false);
|
||||
|
||||
std::array<uint32_t, 3> elements = { (uint32_t)ggml_nelements(dst), 1, 1 };
|
||||
|
||||
ggml_vk_dispatch_pipeline(ctx, subctx, pipeline, { dst_buf }, pc, elements);
|
||||
}
|
||||
|
||||
static void ggml_vk_sin(ggml_backend_vk_context * ctx, vk_context& subctx, const ggml_tensor * src0, ggml_tensor * dst) {
|
||||
ggml_vk_op_f32(ctx, subctx, src0, nullptr, nullptr, nullptr, dst, GGML_OP_SIN, vk_op_unary_push_constants_init(src0, dst));
|
||||
}
|
||||
|
|
@ -11289,6 +11410,12 @@ static bool ggml_vk_build_graph(ggml_backend_vk_context * ctx, ggml_cgraph * cgr
|
|||
case GGML_UNARY_OP_HARDSIGMOID:
|
||||
case GGML_UNARY_OP_HARDSWISH:
|
||||
case GGML_UNARY_OP_ABS:
|
||||
case GGML_UNARY_OP_SOFTPLUS:
|
||||
case GGML_UNARY_OP_STEP:
|
||||
case GGML_UNARY_OP_ROUND:
|
||||
case GGML_UNARY_OP_CEIL:
|
||||
case GGML_UNARY_OP_FLOOR:
|
||||
case GGML_UNARY_OP_TRUNC:
|
||||
break;
|
||||
default:
|
||||
return false;
|
||||
|
|
@ -11330,6 +11457,9 @@ static bool ggml_vk_build_graph(ggml_backend_vk_context * ctx, ggml_cgraph * cgr
|
|||
case GGML_OP_SUB:
|
||||
case GGML_OP_MUL:
|
||||
case GGML_OP_DIV:
|
||||
case GGML_OP_ADD1:
|
||||
case GGML_OP_ARANGE:
|
||||
case GGML_OP_FILL:
|
||||
case GGML_OP_CONCAT:
|
||||
case GGML_OP_UPSCALE:
|
||||
case GGML_OP_SCALE:
|
||||
|
|
@ -11542,6 +11672,18 @@ static bool ggml_vk_build_graph(ggml_backend_vk_context * ctx, ggml_cgraph * cgr
|
|||
case GGML_OP_UPSCALE:
|
||||
ggml_vk_upscale(ctx, compute_ctx, src0, node);
|
||||
|
||||
break;
|
||||
case GGML_OP_ADD1:
|
||||
ggml_vk_add1(ctx, compute_ctx, src0, src1, node);
|
||||
|
||||
break;
|
||||
case GGML_OP_ARANGE:
|
||||
ggml_vk_arange(ctx, compute_ctx, node);
|
||||
|
||||
break;
|
||||
case GGML_OP_FILL:
|
||||
ggml_vk_fill(ctx, compute_ctx, node);
|
||||
|
||||
break;
|
||||
case GGML_OP_SCALE:
|
||||
ggml_vk_scale(ctx, compute_ctx, src0, node);
|
||||
|
|
@ -11626,6 +11768,12 @@ static bool ggml_vk_build_graph(ggml_backend_vk_context * ctx, ggml_cgraph * cgr
|
|||
case GGML_UNARY_OP_HARDSIGMOID:
|
||||
case GGML_UNARY_OP_HARDSWISH:
|
||||
case GGML_UNARY_OP_ABS:
|
||||
case GGML_UNARY_OP_SOFTPLUS:
|
||||
case GGML_UNARY_OP_STEP:
|
||||
case GGML_UNARY_OP_ROUND:
|
||||
case GGML_UNARY_OP_CEIL:
|
||||
case GGML_UNARY_OP_FLOOR:
|
||||
case GGML_UNARY_OP_TRUNC:
|
||||
ggml_vk_unary(ctx, compute_ctx, src0, node);
|
||||
break;
|
||||
default:
|
||||
|
|
@ -11828,6 +11976,9 @@ static bool ggml_vk_compute_forward(ggml_backend_vk_context * ctx, ggml_cgraph *
|
|||
case GGML_OP_SUB:
|
||||
case GGML_OP_MUL:
|
||||
case GGML_OP_DIV:
|
||||
case GGML_OP_ADD1:
|
||||
case GGML_OP_ARANGE:
|
||||
case GGML_OP_FILL:
|
||||
case GGML_OP_ADD_ID:
|
||||
case GGML_OP_CONCAT:
|
||||
case GGML_OP_UPSCALE:
|
||||
|
|
@ -11899,6 +12050,12 @@ static bool ggml_vk_compute_forward(ggml_backend_vk_context * ctx, ggml_cgraph *
|
|||
case GGML_UNARY_OP_HARDSIGMOID:
|
||||
case GGML_UNARY_OP_HARDSWISH:
|
||||
case GGML_UNARY_OP_ABS:
|
||||
case GGML_UNARY_OP_SOFTPLUS:
|
||||
case GGML_UNARY_OP_STEP:
|
||||
case GGML_UNARY_OP_ROUND:
|
||||
case GGML_UNARY_OP_CEIL:
|
||||
case GGML_UNARY_OP_FLOOR:
|
||||
case GGML_UNARY_OP_TRUNC:
|
||||
buf = tensor->buffer;
|
||||
break;
|
||||
default:
|
||||
|
|
@ -13501,6 +13658,12 @@ static bool ggml_backend_vk_device_supports_op(ggml_backend_dev_t dev, const ggm
|
|||
case GGML_UNARY_OP_HARDSIGMOID:
|
||||
case GGML_UNARY_OP_HARDSWISH:
|
||||
case GGML_UNARY_OP_ABS:
|
||||
case GGML_UNARY_OP_SOFTPLUS:
|
||||
case GGML_UNARY_OP_STEP:
|
||||
case GGML_UNARY_OP_ROUND:
|
||||
case GGML_UNARY_OP_CEIL:
|
||||
case GGML_UNARY_OP_FLOOR:
|
||||
case GGML_UNARY_OP_TRUNC:
|
||||
return ggml_is_contiguous(op->src[0]) &&
|
||||
(op->src[0]->type == GGML_TYPE_F32 || op->src[0]->type == GGML_TYPE_F16) &&
|
||||
(op->type == GGML_TYPE_F32 || op->type == GGML_TYPE_F16) &&
|
||||
|
|
@ -13818,6 +13981,9 @@ static bool ggml_backend_vk_device_supports_op(ggml_backend_dev_t dev, const ggm
|
|||
case GGML_OP_UPSCALE:
|
||||
case GGML_OP_ACC:
|
||||
case GGML_OP_CONCAT:
|
||||
case GGML_OP_ADD1:
|
||||
case GGML_OP_ARANGE:
|
||||
case GGML_OP_FILL:
|
||||
case GGML_OP_SCALE:
|
||||
case GGML_OP_PAD:
|
||||
case GGML_OP_ROLL:
|
||||
|
|
@ -14300,6 +14466,16 @@ static void ggml_vk_check_results_0(ggml_backend_vk_context * ctx, ggml_cgraph *
|
|||
} else if (tensor->op == GGML_OP_SCALE) {
|
||||
const float * params = (const float *)tensor->op_params;
|
||||
tensor_clone = ggml_scale_bias(ggml_ctx, src_clone[0], params[0], params[1]);
|
||||
} else if (tensor->op == GGML_OP_ADD1) {
|
||||
tensor_clone = ggml_add1(ggml_ctx, src_clone[0], src_clone[1]);
|
||||
} else if (tensor->op == GGML_OP_ARANGE) {
|
||||
const float start = ggml_get_op_params_f32(tensor, 0);
|
||||
const float stop = ggml_get_op_params_f32(tensor, 1);
|
||||
const float step = ggml_get_op_params_f32(tensor, 2);
|
||||
tensor_clone = ggml_arange(ggml_ctx, start, stop, step);
|
||||
} else if (tensor->op == GGML_OP_FILL) {
|
||||
const float value = ggml_get_op_params_f32(tensor, 0);
|
||||
tensor_clone = ggml_fill(ggml_ctx, tensor_clone, value);
|
||||
} else if (tensor->op == GGML_OP_SQR) {
|
||||
tensor_clone = ggml_sqr(ggml_ctx, src_clone[0]);
|
||||
} else if (tensor->op == GGML_OP_SQRT) {
|
||||
|
|
@ -14413,6 +14589,24 @@ static void ggml_vk_check_results_0(ggml_backend_vk_context * ctx, ggml_cgraph *
|
|||
case GGML_UNARY_OP_ABS:
|
||||
tensor_clone = ggml_abs(ggml_ctx, src_clone[0]);
|
||||
break;
|
||||
case GGML_UNARY_OP_SOFTPLUS:
|
||||
tensor_clone = ggml_softplus(ggml_ctx, src_clone[0]);
|
||||
break;
|
||||
case GGML_UNARY_OP_STEP:
|
||||
tensor_clone = ggml_step(ggml_ctx, src_clone[0]);
|
||||
break;
|
||||
case GGML_UNARY_OP_ROUND:
|
||||
tensor_clone = ggml_round(ggml_ctx, src_clone[0]);
|
||||
break;
|
||||
case GGML_UNARY_OP_CEIL:
|
||||
tensor_clone = ggml_ceil(ggml_ctx, src_clone[0]);
|
||||
break;
|
||||
case GGML_UNARY_OP_FLOOR:
|
||||
tensor_clone = ggml_floor(ggml_ctx, src_clone[0]);
|
||||
break;
|
||||
case GGML_UNARY_OP_TRUNC:
|
||||
tensor_clone = ggml_trunc(ggml_ctx, src_clone[0]);
|
||||
break;
|
||||
default:
|
||||
std::cerr << "Missing vk_check_results OP: " << ggml_op_name(tensor->op) << std::endl;
|
||||
GGML_ABORT("fatal error");
|
||||
|
|
|
|||
|
|
@ -0,0 +1,28 @@
|
|||
#version 450
|
||||
|
||||
#extension GL_EXT_shader_16bit_storage : require
|
||||
|
||||
#include "types.glsl"
|
||||
#include "generic_binary_head.glsl"
|
||||
|
||||
const uint num_threads = 256;
|
||||
|
||||
layout(local_size_x = num_threads, local_size_y = 1, local_size_z = 1) in;
|
||||
|
||||
void main() {
|
||||
uint idx = get_idx();
|
||||
|
||||
const uint num_iter = 2;
|
||||
|
||||
[[unroll]] for (uint i = 0; i < num_iter; ++i) {
|
||||
if (idx >= p.ne) {
|
||||
continue;
|
||||
}
|
||||
uint i00, i01, i02, i03;
|
||||
get_indices(idx, i00, i01, i02, i03);
|
||||
|
||||
data_d[get_doffset() + dst_idx(i00, i01, i02, i03)] = D_TYPE(FLOAT_TYPE(data_a[get_aoffset() + src0_idx(i00, i01, i02, i03)]) + FLOAT_TYPE(data_b[get_boffset()]));
|
||||
|
||||
idx += num_threads;
|
||||
}
|
||||
}
|
||||
|
|
@ -0,0 +1,20 @@
|
|||
#version 450
|
||||
|
||||
#include "generic_head.glsl"
|
||||
#include "types.glsl"
|
||||
|
||||
layout(local_size_x = 512, local_size_y = 1, local_size_z = 1) in;
|
||||
|
||||
layout (binding = 0) writeonly buffer D {D_TYPE data_d[];};
|
||||
|
||||
void main() {
|
||||
const uint i = gl_GlobalInvocationID.x;
|
||||
|
||||
if (i >= p.KX) {
|
||||
return;
|
||||
}
|
||||
|
||||
// p.param1 = start, p.param2 = step
|
||||
float value = p.param1 + p.param2 * float(i);
|
||||
data_d[i] = D_TYPE(value);
|
||||
}
|
||||
|
|
@ -0,0 +1,22 @@
|
|||
#version 450
|
||||
|
||||
#include "generic_head.glsl"
|
||||
#include "types.glsl"
|
||||
|
||||
#extension GL_EXT_control_flow_attributes : enable
|
||||
|
||||
layout(local_size_x = 512, local_size_y = 1, local_size_z = 1) in;
|
||||
|
||||
layout (binding = 0) readonly buffer X {A_TYPE data_a[];};
|
||||
layout (binding = 1) writeonly buffer D {D_TYPE data_d[];};
|
||||
|
||||
void main() {
|
||||
const uint i = gl_GlobalInvocationID.z * 262144 + gl_GlobalInvocationID.y * 512 + gl_GlobalInvocationID.x;
|
||||
|
||||
if (i >= p.KX) {
|
||||
return;
|
||||
}
|
||||
|
||||
const float x = float(data_a[i]);
|
||||
data_d[i] = D_TYPE(ceil(x));
|
||||
}
|
||||
|
|
@ -0,0 +1,19 @@
|
|||
#version 450
|
||||
|
||||
#include "generic_head.glsl"
|
||||
#include "types.glsl"
|
||||
|
||||
layout(local_size_x = 512, local_size_y = 1, local_size_z = 1) in;
|
||||
|
||||
layout (binding = 0) writeonly buffer D {D_TYPE data_d[];};
|
||||
|
||||
void main() {
|
||||
const uint i = gl_GlobalInvocationID.x;
|
||||
|
||||
if (i >= p.KX) {
|
||||
return;
|
||||
}
|
||||
|
||||
// p.param1 = fill value
|
||||
data_d[i] = D_TYPE(p.param1);
|
||||
}
|
||||
|
|
@ -0,0 +1,22 @@
|
|||
#version 450
|
||||
|
||||
#include "generic_head.glsl"
|
||||
#include "types.glsl"
|
||||
|
||||
#extension GL_EXT_control_flow_attributes : enable
|
||||
|
||||
layout(local_size_x = 512, local_size_y = 1, local_size_z = 1) in;
|
||||
|
||||
layout (binding = 0) readonly buffer X {A_TYPE data_a[];};
|
||||
layout (binding = 1) writeonly buffer D {D_TYPE data_d[];};
|
||||
|
||||
void main() {
|
||||
const uint i = gl_GlobalInvocationID.z * 262144 + gl_GlobalInvocationID.y * 512 + gl_GlobalInvocationID.x;
|
||||
|
||||
if (i >= p.KX) {
|
||||
return;
|
||||
}
|
||||
|
||||
const float x = float(data_a[i]);
|
||||
data_d[i] = D_TYPE(floor(x));
|
||||
}
|
||||
|
|
@ -0,0 +1,29 @@
|
|||
#version 450
|
||||
|
||||
#include "generic_head.glsl"
|
||||
#include "types.glsl"
|
||||
|
||||
#extension GL_EXT_control_flow_attributes : enable
|
||||
|
||||
layout(local_size_x = 512, local_size_y = 1, local_size_z = 1) in;
|
||||
|
||||
layout (binding = 0) readonly buffer X {A_TYPE data_a[];};
|
||||
layout (binding = 1) writeonly buffer D {D_TYPE data_d[];};
|
||||
|
||||
void main() {
|
||||
const uint i = gl_GlobalInvocationID.z * 262144 + gl_GlobalInvocationID.y * 512 + gl_GlobalInvocationID.x;
|
||||
|
||||
if (i >= p.KX) {
|
||||
return;
|
||||
}
|
||||
|
||||
const float x = float(data_a[i]);
|
||||
float result;
|
||||
// Round halfway cases away from zero as roundf does.
|
||||
if (x >= 0.0) {
|
||||
result = floor(x + 0.5);
|
||||
} else {
|
||||
result = ceil(x - 0.5);
|
||||
}
|
||||
data_d[i] = D_TYPE(result);
|
||||
}
|
||||
|
|
@ -0,0 +1,23 @@
|
|||
#version 450
|
||||
|
||||
#include "generic_head.glsl"
|
||||
#include "types.glsl"
|
||||
|
||||
#extension GL_EXT_control_flow_attributes : enable
|
||||
|
||||
layout(local_size_x = 512, local_size_y = 1, local_size_z = 1) in;
|
||||
|
||||
layout (binding = 0) readonly buffer X {A_TYPE data_a[];};
|
||||
layout (binding = 1) writeonly buffer D {D_TYPE data_d[];};
|
||||
|
||||
void main() {
|
||||
const uint i = gl_GlobalInvocationID.z * 262144 + gl_GlobalInvocationID.y * 512 + gl_GlobalInvocationID.x;
|
||||
|
||||
if (i >= p.KX) {
|
||||
return;
|
||||
}
|
||||
|
||||
const float x = float(data_a[i]);
|
||||
const float result = (x > 20.0f) ? x : log(1.0f + exp(x));
|
||||
data_d[i] = D_TYPE(result);
|
||||
}
|
||||
|
|
@ -0,0 +1,22 @@
|
|||
#version 450
|
||||
|
||||
#include "generic_head.glsl"
|
||||
#include "types.glsl"
|
||||
|
||||
#extension GL_EXT_control_flow_attributes : enable
|
||||
|
||||
layout(local_size_x = 512, local_size_y = 1, local_size_z = 1) in;
|
||||
|
||||
layout (binding = 0) readonly buffer X {A_TYPE data_a[];};
|
||||
layout (binding = 1) writeonly buffer D {D_TYPE data_d[];};
|
||||
|
||||
void main() {
|
||||
const uint i = gl_GlobalInvocationID.z * 262144 + gl_GlobalInvocationID.y * 512 + gl_GlobalInvocationID.x;
|
||||
|
||||
if (i >= p.KX) {
|
||||
return;
|
||||
}
|
||||
|
||||
const float x = float(data_a[i]);
|
||||
data_d[i] = D_TYPE(x >= 0.0f ? 1.0f : 0.0f);
|
||||
}
|
||||
|
|
@ -0,0 +1,22 @@
|
|||
#version 450
|
||||
|
||||
#include "generic_head.glsl"
|
||||
#include "types.glsl"
|
||||
|
||||
#extension GL_EXT_control_flow_attributes : enable
|
||||
|
||||
layout(local_size_x = 512, local_size_y = 1, local_size_z = 1) in;
|
||||
|
||||
layout (binding = 0) readonly buffer X {A_TYPE data_a[];};
|
||||
layout (binding = 1) writeonly buffer D {D_TYPE data_d[];};
|
||||
|
||||
void main() {
|
||||
const uint i = gl_GlobalInvocationID.z * 262144 + gl_GlobalInvocationID.y * 512 + gl_GlobalInvocationID.x;
|
||||
|
||||
if (i >= p.KX) {
|
||||
return;
|
||||
}
|
||||
|
||||
const float x = float(data_a[i]);
|
||||
data_d[i] = D_TYPE(trunc(x));
|
||||
}
|
||||
|
|
@ -846,6 +846,25 @@ void process_shaders() {
|
|||
string_to_spv("abs_f16", "abs.comp", {{"A_TYPE", "float16_t"}, {"D_TYPE", "float16_t"}});
|
||||
string_to_spv("abs_f32", "abs.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}});
|
||||
|
||||
string_to_spv("softplus_f16", "softplus.comp", {{"A_TYPE", "float16_t"}, {"D_TYPE", "float16_t"}});
|
||||
string_to_spv("softplus_f32", "softplus.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}});
|
||||
|
||||
string_to_spv("add1_f16_f16", "add1.comp", {{"A_TYPE", "float16_t"}, {"B_TYPE", "float16_t"}, {"D_TYPE", "float16_t"}, {"FLOAT_TYPE", "float"}});
|
||||
string_to_spv("add1_f16_f32", "add1.comp", {{"A_TYPE", "float16_t"}, {"B_TYPE", "float"}, {"D_TYPE", "float16_t"}, {"FLOAT_TYPE", "float"}});
|
||||
string_to_spv("add1_f32_f32", "add1.comp", {{"A_TYPE", "float"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}});
|
||||
string_to_spv("arange_f32", "arange.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}});
|
||||
string_to_spv("fill_f32", "fill.comp", {{"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}});
|
||||
string_to_spv("step_f16", "step.comp", {{"A_TYPE", "float16_t"}, {"D_TYPE", "float16_t"}});
|
||||
string_to_spv("step_f32", "step.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}});
|
||||
string_to_spv("round_f16", "round.comp", {{"A_TYPE", "float16_t"}, {"D_TYPE", "float16_t"}});
|
||||
string_to_spv("round_f32", "round.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}});
|
||||
string_to_spv("ceil_f16", "ceil.comp", {{"A_TYPE", "float16_t"}, {"D_TYPE", "float16_t"}});
|
||||
string_to_spv("ceil_f32", "ceil.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}});
|
||||
string_to_spv("floor_f16", "floor.comp", {{"A_TYPE", "float16_t"}, {"D_TYPE", "float16_t"}});
|
||||
string_to_spv("floor_f32", "floor.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}});
|
||||
string_to_spv("trunc_f16", "trunc.comp", {{"A_TYPE", "float16_t"}, {"D_TYPE", "float16_t"}});
|
||||
string_to_spv("trunc_f32", "trunc.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}});
|
||||
|
||||
for (auto rte : {false, true}) {
|
||||
std::string suffix = rte ? "_rte" : "";
|
||||
string_to_spv("geglu_f16" + suffix, "geglu.comp", {{"A_TYPE", "float16_t"}, {"D_TYPE", "float16_t"}, {"RTE16", rte ? "1" : "0"}});
|
||||
|
|
|
|||
|
|
@ -7026,6 +7026,7 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
|
|||
test_cases.emplace_back(new test_bin_bcast(ggml_add, GGML_TYPE_F32, {16, 5, 4, 3}, {1, 1, 1, 1}, 16));
|
||||
|
||||
test_cases.emplace_back(new test_add1());
|
||||
test_cases.emplace_back(new test_add1(GGML_TYPE_F32, {1024, 1024, 1, 1}));
|
||||
test_cases.emplace_back(new test_scale());
|
||||
test_cases.emplace_back(new test_scale(GGML_TYPE_F32, {10, 10, 10, 10}, 2.0f, 1.0f));
|
||||
test_cases.emplace_back(new test_scale(GGML_TYPE_F32, {10, 10, 10, 10}, 2.0f, 1.0f, true)); // inplace test
|
||||
|
|
@ -7365,9 +7366,13 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
|
|||
test_cases.emplace_back(new test_clamp (type, {7, 1, 5, 3}));
|
||||
test_cases.emplace_back(new test_leaky_relu(type, {7, 1, 5, 3}));
|
||||
test_cases.emplace_back(new test_floor (type, {7, 1, 5, 3}));
|
||||
test_cases.emplace_back(new test_floor (type, { 1024, 1024, 1, 1 }));
|
||||
test_cases.emplace_back(new test_ceil (type, {7, 1, 5, 3}));
|
||||
test_cases.emplace_back(new test_ceil (type, { 1024, 1024, 1, 1 }));
|
||||
test_cases.emplace_back(new test_round (type, {7, 1, 5, 3}));
|
||||
test_cases.emplace_back(new test_round (type, { 1024, 1024, 1, 1 }));
|
||||
test_cases.emplace_back(new test_trunc (type, {7, 1, 5, 3}));
|
||||
test_cases.emplace_back(new test_trunc (type, { 1024, 1024, 1, 1 }));
|
||||
}
|
||||
|
||||
test_cases.emplace_back(new test_diag_mask_inf(GGML_TYPE_F32, {10, 10, 1, 1}, 5));
|
||||
|
|
@ -7569,6 +7574,7 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
|
|||
test_cases.emplace_back(new test_pad_reflect_1d(GGML_TYPE_F32, {3000, 384, 4, 1}));
|
||||
test_cases.emplace_back(new test_roll());
|
||||
test_cases.emplace_back(new test_arange());
|
||||
test_cases.emplace_back(new test_arange(GGML_TYPE_F32, 0.0f, 1048576.0f, 1.0f));
|
||||
test_cases.emplace_back(new test_timestep_embedding());
|
||||
test_cases.emplace_back(new test_leaky_relu());
|
||||
|
||||
|
|
@ -7596,6 +7602,7 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
|
|||
test_cases.emplace_back(new test_fill(0.0f));
|
||||
test_cases.emplace_back(new test_fill(2.0f, GGML_TYPE_F32, { 303, 207, 11, 3 }));
|
||||
test_cases.emplace_back(new test_fill(-152.0f, GGML_TYPE_F32, { 800, 600, 4, 4 }));
|
||||
test_cases.emplace_back(new test_fill(3.5f, GGML_TYPE_F32, { 2048, 512, 2, 2 }));
|
||||
|
||||
test_cases.emplace_back(new test_solve_tri());
|
||||
test_cases.emplace_back(new test_solve_tri(GGML_TYPE_F32, { 11, 11, 1, 1 }, { 5, 11, 1, 1 }));
|
||||
|
|
|
|||
Loading…
Reference in New Issue