Added capabilities for specialization data based on the size of the vectors passed
This commit is contained in:
parent
b80548ac3d
commit
23cf43e231
5 changed files with 140 additions and 109 deletions
|
|
@ -12,6 +12,10 @@ layout(set = 0, binding = 2) buffer tensorOutput {
|
|||
uint valuesOutput[ ];
|
||||
};
|
||||
|
||||
layout (constant_id = 0) const uint LEN_LHS = 0;
|
||||
layout (constant_id = 1) const uint LEN_RHS = 0;
|
||||
layout (constant_id = 2) const uint LEN_OUT = 0;
|
||||
|
||||
// TODO: Explore how to make layout inside shader dynamic
|
||||
layout (local_size_x = 1, local_size_y = 1, local_size_z = 1) in;
|
||||
|
||||
|
|
|
|||
|
|
@ -36,7 +36,7 @@ namespace kp {
|
|||
namespace shader_data {
|
||||
static unsigned const char shaders_glsl_opmult_comp_spv[] = {
|
||||
0x03, 0x02, 0x23, 0x07, 0x00, 0x00, 0x01, 0x00, 0x08, 0x00, 0x08, 0x00,
|
||||
0x2a, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x11, 0x00, 0x02, 0x00,
|
||||
0x26, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x11, 0x00, 0x02, 0x00,
|
||||
0x01, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x06, 0x00, 0x01, 0x00, 0x00, 0x00,
|
||||
0x47, 0x4c, 0x53, 0x4c, 0x2e, 0x73, 0x74, 0x64, 0x2e, 0x34, 0x35, 0x30,
|
||||
0x00, 0x00, 0x00, 0x00, 0x0e, 0x00, 0x03, 0x00, 0x00, 0x00, 0x00, 0x00,
|
||||
|
|
@ -56,16 +56,20 @@ static unsigned const char shaders_glsl_opmult_comp_spv[] = {
|
|||
0x11, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x76, 0x61, 0x6c, 0x75,
|
||||
0x65, 0x73, 0x4f, 0x75, 0x74, 0x70, 0x75, 0x74, 0x00, 0x00, 0x00, 0x00,
|
||||
0x05, 0x00, 0x03, 0x00, 0x13, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
|
||||
0x05, 0x00, 0x05, 0x00, 0x18, 0x00, 0x00, 0x00, 0x74, 0x65, 0x6e, 0x73,
|
||||
0x6f, 0x72, 0x4c, 0x68, 0x73, 0x00, 0x00, 0x00, 0x06, 0x00, 0x06, 0x00,
|
||||
0x18, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x76, 0x61, 0x6c, 0x75,
|
||||
0x65, 0x73, 0x4c, 0x68, 0x73, 0x00, 0x00, 0x00, 0x05, 0x00, 0x03, 0x00,
|
||||
0x1a, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x05, 0x00, 0x05, 0x00,
|
||||
0x20, 0x00, 0x00, 0x00, 0x74, 0x65, 0x6e, 0x73, 0x6f, 0x72, 0x52, 0x68,
|
||||
0x73, 0x00, 0x00, 0x00, 0x06, 0x00, 0x06, 0x00, 0x20, 0x00, 0x00, 0x00,
|
||||
0x00, 0x00, 0x00, 0x00, 0x76, 0x61, 0x6c, 0x75, 0x65, 0x73, 0x52, 0x68,
|
||||
0x73, 0x00, 0x00, 0x00, 0x05, 0x00, 0x03, 0x00, 0x22, 0x00, 0x00, 0x00,
|
||||
0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x0b, 0x00, 0x00, 0x00,
|
||||
0x05, 0x00, 0x04, 0x00, 0x17, 0x00, 0x00, 0x00, 0x4c, 0x45, 0x4e, 0x5f,
|
||||
0x52, 0x48, 0x53, 0x00, 0x05, 0x00, 0x05, 0x00, 0x1b, 0x00, 0x00, 0x00,
|
||||
0x74, 0x65, 0x6e, 0x73, 0x6f, 0x72, 0x4c, 0x68, 0x73, 0x00, 0x00, 0x00,
|
||||
0x06, 0x00, 0x06, 0x00, 0x1b, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
|
||||
0x76, 0x61, 0x6c, 0x75, 0x65, 0x73, 0x4c, 0x68, 0x73, 0x00, 0x00, 0x00,
|
||||
0x05, 0x00, 0x03, 0x00, 0x1d, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
|
||||
0x05, 0x00, 0x05, 0x00, 0x1f, 0x00, 0x00, 0x00, 0x74, 0x65, 0x6e, 0x73,
|
||||
0x6f, 0x72, 0x52, 0x68, 0x73, 0x00, 0x00, 0x00, 0x06, 0x00, 0x06, 0x00,
|
||||
0x1f, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x76, 0x61, 0x6c, 0x75,
|
||||
0x65, 0x73, 0x52, 0x68, 0x73, 0x00, 0x00, 0x00, 0x05, 0x00, 0x03, 0x00,
|
||||
0x21, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x05, 0x00, 0x04, 0x00,
|
||||
0x22, 0x00, 0x00, 0x00, 0x4c, 0x45, 0x4e, 0x5f, 0x4c, 0x48, 0x53, 0x00,
|
||||
0x05, 0x00, 0x04, 0x00, 0x23, 0x00, 0x00, 0x00, 0x4c, 0x45, 0x4e, 0x5f,
|
||||
0x4f, 0x55, 0x54, 0x00, 0x47, 0x00, 0x04, 0x00, 0x0b, 0x00, 0x00, 0x00,
|
||||
0x0b, 0x00, 0x00, 0x00, 0x1c, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00,
|
||||
0x10, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00,
|
||||
0x48, 0x00, 0x05, 0x00, 0x11, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
|
||||
|
|
@ -74,20 +78,24 @@ static unsigned const char shaders_glsl_opmult_comp_spv[] = {
|
|||
0x13, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
|
||||
0x47, 0x00, 0x04, 0x00, 0x13, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00,
|
||||
0x02, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x17, 0x00, 0x00, 0x00,
|
||||
0x01, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00,
|
||||
0x1a, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00,
|
||||
0x48, 0x00, 0x05, 0x00, 0x1b, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
|
||||
0x23, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x03, 0x00,
|
||||
0x1b, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00,
|
||||
0x1d, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
|
||||
0x47, 0x00, 0x04, 0x00, 0x1d, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00,
|
||||
0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x1e, 0x00, 0x00, 0x00,
|
||||
0x06, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x48, 0x00, 0x05, 0x00,
|
||||
0x18, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00,
|
||||
0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x03, 0x00, 0x18, 0x00, 0x00, 0x00,
|
||||
0x03, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x1a, 0x00, 0x00, 0x00,
|
||||
0x1f, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00,
|
||||
0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x03, 0x00, 0x1f, 0x00, 0x00, 0x00,
|
||||
0x03, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x21, 0x00, 0x00, 0x00,
|
||||
0x22, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00,
|
||||
0x1a, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
|
||||
0x47, 0x00, 0x04, 0x00, 0x1f, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00,
|
||||
0x04, 0x00, 0x00, 0x00, 0x48, 0x00, 0x05, 0x00, 0x20, 0x00, 0x00, 0x00,
|
||||
0x00, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
|
||||
0x47, 0x00, 0x03, 0x00, 0x20, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00,
|
||||
0x47, 0x00, 0x04, 0x00, 0x22, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00,
|
||||
0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x22, 0x00, 0x00, 0x00,
|
||||
0x21, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00,
|
||||
0x29, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x19, 0x00, 0x00, 0x00,
|
||||
0x21, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
|
||||
0x47, 0x00, 0x04, 0x00, 0x22, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
|
||||
0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x23, 0x00, 0x00, 0x00,
|
||||
0x01, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00,
|
||||
0x25, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x19, 0x00, 0x00, 0x00,
|
||||
0x13, 0x00, 0x02, 0x00, 0x02, 0x00, 0x00, 0x00, 0x21, 0x00, 0x03, 0x00,
|
||||
0x03, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x15, 0x00, 0x04, 0x00,
|
||||
0x06, 0x00, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
|
||||
|
|
@ -106,21 +114,25 @@ static unsigned const char shaders_glsl_opmult_comp_spv[] = {
|
|||
0x13, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x15, 0x00, 0x04, 0x00,
|
||||
0x14, 0x00, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
|
||||
0x2b, 0x00, 0x04, 0x00, 0x14, 0x00, 0x00, 0x00, 0x15, 0x00, 0x00, 0x00,
|
||||
0x00, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x03, 0x00, 0x17, 0x00, 0x00, 0x00,
|
||||
0x06, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x03, 0x00, 0x18, 0x00, 0x00, 0x00,
|
||||
0x17, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x19, 0x00, 0x00, 0x00,
|
||||
0x02, 0x00, 0x00, 0x00, 0x18, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00,
|
||||
0x19, 0x00, 0x00, 0x00, 0x1a, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
|
||||
0x00, 0x00, 0x00, 0x00, 0x32, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
|
||||
0x17, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00,
|
||||
0x18, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00,
|
||||
0x1d, 0x00, 0x03, 0x00, 0x1a, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00,
|
||||
0x1e, 0x00, 0x03, 0x00, 0x1b, 0x00, 0x00, 0x00, 0x1a, 0x00, 0x00, 0x00,
|
||||
0x20, 0x00, 0x04, 0x00, 0x1c, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
|
||||
0x06, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x03, 0x00, 0x1f, 0x00, 0x00, 0x00,
|
||||
0x06, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x03, 0x00, 0x20, 0x00, 0x00, 0x00,
|
||||
0x1f, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x21, 0x00, 0x00, 0x00,
|
||||
0x02, 0x00, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00,
|
||||
0x21, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
|
||||
0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x28, 0x00, 0x00, 0x00,
|
||||
0x1b, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x1c, 0x00, 0x00, 0x00,
|
||||
0x1d, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x03, 0x00,
|
||||
0x1e, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x03, 0x00,
|
||||
0x1f, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00,
|
||||
0x20, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x1f, 0x00, 0x00, 0x00,
|
||||
0x3b, 0x00, 0x04, 0x00, 0x20, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00,
|
||||
0x02, 0x00, 0x00, 0x00, 0x32, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
|
||||
0x22, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x32, 0x00, 0x04, 0x00,
|
||||
0x06, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
|
||||
0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x24, 0x00, 0x00, 0x00,
|
||||
0x01, 0x00, 0x00, 0x00, 0x2c, 0x00, 0x06, 0x00, 0x09, 0x00, 0x00, 0x00,
|
||||
0x29, 0x00, 0x00, 0x00, 0x28, 0x00, 0x00, 0x00, 0x28, 0x00, 0x00, 0x00,
|
||||
0x28, 0x00, 0x00, 0x00, 0x36, 0x00, 0x05, 0x00, 0x02, 0x00, 0x00, 0x00,
|
||||
0x25, 0x00, 0x00, 0x00, 0x24, 0x00, 0x00, 0x00, 0x24, 0x00, 0x00, 0x00,
|
||||
0x24, 0x00, 0x00, 0x00, 0x36, 0x00, 0x05, 0x00, 0x02, 0x00, 0x00, 0x00,
|
||||
0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00,
|
||||
0xf8, 0x00, 0x02, 0x00, 0x05, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00,
|
||||
0x07, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00,
|
||||
|
|
@ -129,23 +141,12 @@ static unsigned const char shaders_glsl_opmult_comp_spv[] = {
|
|||
0x06, 0x00, 0x00, 0x00, 0x0f, 0x00, 0x00, 0x00, 0x0e, 0x00, 0x00, 0x00,
|
||||
0x3e, 0x00, 0x03, 0x00, 0x08, 0x00, 0x00, 0x00, 0x0f, 0x00, 0x00, 0x00,
|
||||
0x3d, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x16, 0x00, 0x00, 0x00,
|
||||
0x08, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
|
||||
0x1b, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x41, 0x00, 0x06, 0x00,
|
||||
0x1c, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x00, 0x00, 0x1a, 0x00, 0x00, 0x00,
|
||||
0x15, 0x00, 0x00, 0x00, 0x1b, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00,
|
||||
0x06, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x00, 0x00,
|
||||
0x3d, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00,
|
||||
0x08, 0x00, 0x00, 0x00, 0x41, 0x00, 0x06, 0x00, 0x1c, 0x00, 0x00, 0x00,
|
||||
0x24, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0x15, 0x00, 0x00, 0x00,
|
||||
0x23, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
|
||||
0x25, 0x00, 0x00, 0x00, 0x24, 0x00, 0x00, 0x00, 0x84, 0x00, 0x05, 0x00,
|
||||
0x06, 0x00, 0x00, 0x00, 0x26, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x00, 0x00,
|
||||
0x25, 0x00, 0x00, 0x00, 0x41, 0x00, 0x06, 0x00, 0x1c, 0x00, 0x00, 0x00,
|
||||
0x27, 0x00, 0x00, 0x00, 0x13, 0x00, 0x00, 0x00, 0x15, 0x00, 0x00, 0x00,
|
||||
0x16, 0x00, 0x00, 0x00, 0x3e, 0x00, 0x03, 0x00, 0x27, 0x00, 0x00, 0x00,
|
||||
0x26, 0x00, 0x00, 0x00, 0xfd, 0x00, 0x01, 0x00, 0x38, 0x00, 0x01, 0x00
|
||||
0x08, 0x00, 0x00, 0x00, 0x41, 0x00, 0x06, 0x00, 0x18, 0x00, 0x00, 0x00,
|
||||
0x19, 0x00, 0x00, 0x00, 0x13, 0x00, 0x00, 0x00, 0x15, 0x00, 0x00, 0x00,
|
||||
0x16, 0x00, 0x00, 0x00, 0x3e, 0x00, 0x03, 0x00, 0x19, 0x00, 0x00, 0x00,
|
||||
0x17, 0x00, 0x00, 0x00, 0xfd, 0x00, 0x01, 0x00, 0x38, 0x00, 0x01, 0x00
|
||||
};
|
||||
static unsigned const int shaders_glsl_opmult_comp_spv_len = 1308;
|
||||
static unsigned const int shaders_glsl_opmult_comp_spv_len = 1320;
|
||||
}
|
||||
}
|
||||
#endif // define SHADEROP_SHADEROPMULT_HPP
|
||||
|
|
@ -514,6 +515,7 @@ class Sequence
|
|||
* not be able to add the operation.
|
||||
*
|
||||
* @param tensors Vector of tensors to use for the operation
|
||||
* @param TArgs Template parameters that are used to initialise operation which allows for extensible configurations on initialisation.
|
||||
*/
|
||||
template<typename T, typename... TArgs>
|
||||
bool record(std::vector<std::shared_ptr<Tensor>> tensors, TArgs&&... params)
|
||||
|
|
@ -642,6 +644,7 @@ class Manager
|
|||
*
|
||||
* @param tensors The tensors to be used in the operation recorded
|
||||
* @param sequenceName The name of the sequence to be retrieved or created
|
||||
* @param TArgs Template parameters that will be used to initialise Operation to allow for extensible configurations on initialisation
|
||||
*/
|
||||
template<typename T, typename... TArgs>
|
||||
void evalOp(std::vector<std::shared_ptr<Tensor>> tensors,
|
||||
|
|
@ -773,7 +776,7 @@ class Algorithm
|
|||
|
||||
// Create util functions
|
||||
void createShaderModule(const std::vector<char>& shaderFileData);
|
||||
void createPipeline();
|
||||
void createPipeline(std::vector<uint32_t> specializationData = {});
|
||||
|
||||
// Parameters
|
||||
void createParameters(std::vector<std::shared_ptr<Tensor>>& tensorParams);
|
||||
|
|
|
|||
|
|
@ -38,7 +38,13 @@ Algorithm::init(const std::vector<char>& shaderFileData,
|
|||
// TODO: Move to util function
|
||||
this->createParameters(tensorParams);
|
||||
this->createShaderModule(shaderFileData);
|
||||
this->createPipeline();
|
||||
|
||||
std::vector<uint32_t> sizes;
|
||||
for (std::shared_ptr<Tensor> tensor: tensorParams) {
|
||||
SPDLOG_WARN("size: {}", tensor->size());
|
||||
sizes.push_back(tensor->size());
|
||||
}
|
||||
this->createPipeline(sizes);
|
||||
}
|
||||
|
||||
void
|
||||
|
|
@ -152,7 +158,7 @@ Algorithm::createShaderModule(const std::vector<char>& shaderFileData)
|
|||
}
|
||||
|
||||
void
|
||||
Algorithm::createPipeline()
|
||||
Algorithm::createPipeline(std::vector<uint32_t> specializationData)
|
||||
{
|
||||
SPDLOG_DEBUG("Kompute Algorithm calling create Pipeline");
|
||||
|
||||
|
|
@ -166,12 +172,29 @@ Algorithm::createPipeline()
|
|||
this->mDevice->createPipelineLayout(
|
||||
&pipelineLayoutInfo, nullptr, this->mPipelineLayout.get());
|
||||
|
||||
std::vector<vk::SpecializationMapEntry> specializationEntries;
|
||||
|
||||
for (size_t i = 0; i < specializationData.size(); i++) {
|
||||
vk::SpecializationMapEntry specializationEntry(
|
||||
static_cast<uint32_t>(i),
|
||||
static_cast<uint32_t>(sizeof(uint32_t) * i),
|
||||
sizeof(uint32_t));
|
||||
|
||||
specializationEntries.push_back(specializationEntry);
|
||||
}
|
||||
|
||||
vk::SpecializationInfo specializationInfo(
|
||||
static_cast<uint32_t>(specializationEntries.size()),
|
||||
specializationEntries.data(),
|
||||
sizeof(uint32_t) * specializationEntries.size(),
|
||||
specializationData.data());
|
||||
|
||||
vk::PipelineShaderStageCreateInfo shaderStage(
|
||||
vk::PipelineShaderStageCreateFlags(),
|
||||
vk::ShaderStageFlagBits::eCompute,
|
||||
*this->mShaderModule,
|
||||
"main",
|
||||
nullptr);
|
||||
&specializationInfo);
|
||||
|
||||
vk::ComputePipelineCreateInfo pipelineInfo(vk::PipelineCreateFlags(),
|
||||
shaderStage,
|
||||
|
|
|
|||
|
|
@ -80,7 +80,7 @@ class Algorithm
|
|||
|
||||
// Create util functions
|
||||
void createShaderModule(const std::vector<char>& shaderFileData);
|
||||
void createPipeline();
|
||||
void createPipeline(std::vector<uint32_t> specializationData = {});
|
||||
|
||||
// Parameters
|
||||
void createParameters(std::vector<std::shared_ptr<Tensor>>& tensorParams);
|
||||
|
|
|
|||
|
|
@ -25,7 +25,7 @@ namespace kp {
|
|||
namespace shader_data {
|
||||
static unsigned const char shaders_glsl_opmult_comp_spv[] = {
|
||||
0x03, 0x02, 0x23, 0x07, 0x00, 0x00, 0x01, 0x00, 0x08, 0x00, 0x08, 0x00,
|
||||
0x2a, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x11, 0x00, 0x02, 0x00,
|
||||
0x26, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x11, 0x00, 0x02, 0x00,
|
||||
0x01, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x06, 0x00, 0x01, 0x00, 0x00, 0x00,
|
||||
0x47, 0x4c, 0x53, 0x4c, 0x2e, 0x73, 0x74, 0x64, 0x2e, 0x34, 0x35, 0x30,
|
||||
0x00, 0x00, 0x00, 0x00, 0x0e, 0x00, 0x03, 0x00, 0x00, 0x00, 0x00, 0x00,
|
||||
|
|
@ -45,16 +45,20 @@ static unsigned const char shaders_glsl_opmult_comp_spv[] = {
|
|||
0x11, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x76, 0x61, 0x6c, 0x75,
|
||||
0x65, 0x73, 0x4f, 0x75, 0x74, 0x70, 0x75, 0x74, 0x00, 0x00, 0x00, 0x00,
|
||||
0x05, 0x00, 0x03, 0x00, 0x13, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
|
||||
0x05, 0x00, 0x05, 0x00, 0x18, 0x00, 0x00, 0x00, 0x74, 0x65, 0x6e, 0x73,
|
||||
0x6f, 0x72, 0x4c, 0x68, 0x73, 0x00, 0x00, 0x00, 0x06, 0x00, 0x06, 0x00,
|
||||
0x18, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x76, 0x61, 0x6c, 0x75,
|
||||
0x65, 0x73, 0x4c, 0x68, 0x73, 0x00, 0x00, 0x00, 0x05, 0x00, 0x03, 0x00,
|
||||
0x1a, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x05, 0x00, 0x05, 0x00,
|
||||
0x20, 0x00, 0x00, 0x00, 0x74, 0x65, 0x6e, 0x73, 0x6f, 0x72, 0x52, 0x68,
|
||||
0x73, 0x00, 0x00, 0x00, 0x06, 0x00, 0x06, 0x00, 0x20, 0x00, 0x00, 0x00,
|
||||
0x00, 0x00, 0x00, 0x00, 0x76, 0x61, 0x6c, 0x75, 0x65, 0x73, 0x52, 0x68,
|
||||
0x73, 0x00, 0x00, 0x00, 0x05, 0x00, 0x03, 0x00, 0x22, 0x00, 0x00, 0x00,
|
||||
0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x0b, 0x00, 0x00, 0x00,
|
||||
0x05, 0x00, 0x04, 0x00, 0x17, 0x00, 0x00, 0x00, 0x4c, 0x45, 0x4e, 0x5f,
|
||||
0x52, 0x48, 0x53, 0x00, 0x05, 0x00, 0x05, 0x00, 0x1b, 0x00, 0x00, 0x00,
|
||||
0x74, 0x65, 0x6e, 0x73, 0x6f, 0x72, 0x4c, 0x68, 0x73, 0x00, 0x00, 0x00,
|
||||
0x06, 0x00, 0x06, 0x00, 0x1b, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
|
||||
0x76, 0x61, 0x6c, 0x75, 0x65, 0x73, 0x4c, 0x68, 0x73, 0x00, 0x00, 0x00,
|
||||
0x05, 0x00, 0x03, 0x00, 0x1d, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
|
||||
0x05, 0x00, 0x05, 0x00, 0x1f, 0x00, 0x00, 0x00, 0x74, 0x65, 0x6e, 0x73,
|
||||
0x6f, 0x72, 0x52, 0x68, 0x73, 0x00, 0x00, 0x00, 0x06, 0x00, 0x06, 0x00,
|
||||
0x1f, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x76, 0x61, 0x6c, 0x75,
|
||||
0x65, 0x73, 0x52, 0x68, 0x73, 0x00, 0x00, 0x00, 0x05, 0x00, 0x03, 0x00,
|
||||
0x21, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x05, 0x00, 0x04, 0x00,
|
||||
0x22, 0x00, 0x00, 0x00, 0x4c, 0x45, 0x4e, 0x5f, 0x4c, 0x48, 0x53, 0x00,
|
||||
0x05, 0x00, 0x04, 0x00, 0x23, 0x00, 0x00, 0x00, 0x4c, 0x45, 0x4e, 0x5f,
|
||||
0x4f, 0x55, 0x54, 0x00, 0x47, 0x00, 0x04, 0x00, 0x0b, 0x00, 0x00, 0x00,
|
||||
0x0b, 0x00, 0x00, 0x00, 0x1c, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00,
|
||||
0x10, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00,
|
||||
0x48, 0x00, 0x05, 0x00, 0x11, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
|
||||
|
|
@ -63,20 +67,24 @@ static unsigned const char shaders_glsl_opmult_comp_spv[] = {
|
|||
0x13, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
|
||||
0x47, 0x00, 0x04, 0x00, 0x13, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00,
|
||||
0x02, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x17, 0x00, 0x00, 0x00,
|
||||
0x01, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00,
|
||||
0x1a, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00,
|
||||
0x48, 0x00, 0x05, 0x00, 0x1b, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
|
||||
0x23, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x03, 0x00,
|
||||
0x1b, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00,
|
||||
0x1d, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
|
||||
0x47, 0x00, 0x04, 0x00, 0x1d, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00,
|
||||
0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x1e, 0x00, 0x00, 0x00,
|
||||
0x06, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x48, 0x00, 0x05, 0x00,
|
||||
0x18, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00,
|
||||
0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x03, 0x00, 0x18, 0x00, 0x00, 0x00,
|
||||
0x03, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x1a, 0x00, 0x00, 0x00,
|
||||
0x1f, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00,
|
||||
0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x03, 0x00, 0x1f, 0x00, 0x00, 0x00,
|
||||
0x03, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x21, 0x00, 0x00, 0x00,
|
||||
0x22, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00,
|
||||
0x1a, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
|
||||
0x47, 0x00, 0x04, 0x00, 0x1f, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00,
|
||||
0x04, 0x00, 0x00, 0x00, 0x48, 0x00, 0x05, 0x00, 0x20, 0x00, 0x00, 0x00,
|
||||
0x00, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
|
||||
0x47, 0x00, 0x03, 0x00, 0x20, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00,
|
||||
0x47, 0x00, 0x04, 0x00, 0x22, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00,
|
||||
0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x22, 0x00, 0x00, 0x00,
|
||||
0x21, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00,
|
||||
0x29, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x19, 0x00, 0x00, 0x00,
|
||||
0x21, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
|
||||
0x47, 0x00, 0x04, 0x00, 0x22, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
|
||||
0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x23, 0x00, 0x00, 0x00,
|
||||
0x01, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00,
|
||||
0x25, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x19, 0x00, 0x00, 0x00,
|
||||
0x13, 0x00, 0x02, 0x00, 0x02, 0x00, 0x00, 0x00, 0x21, 0x00, 0x03, 0x00,
|
||||
0x03, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x15, 0x00, 0x04, 0x00,
|
||||
0x06, 0x00, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
|
||||
|
|
@ -95,21 +103,25 @@ static unsigned const char shaders_glsl_opmult_comp_spv[] = {
|
|||
0x13, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x15, 0x00, 0x04, 0x00,
|
||||
0x14, 0x00, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
|
||||
0x2b, 0x00, 0x04, 0x00, 0x14, 0x00, 0x00, 0x00, 0x15, 0x00, 0x00, 0x00,
|
||||
0x00, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x03, 0x00, 0x17, 0x00, 0x00, 0x00,
|
||||
0x06, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x03, 0x00, 0x18, 0x00, 0x00, 0x00,
|
||||
0x17, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x19, 0x00, 0x00, 0x00,
|
||||
0x02, 0x00, 0x00, 0x00, 0x18, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00,
|
||||
0x19, 0x00, 0x00, 0x00, 0x1a, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
|
||||
0x00, 0x00, 0x00, 0x00, 0x32, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
|
||||
0x17, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00,
|
||||
0x18, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00,
|
||||
0x1d, 0x00, 0x03, 0x00, 0x1a, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00,
|
||||
0x1e, 0x00, 0x03, 0x00, 0x1b, 0x00, 0x00, 0x00, 0x1a, 0x00, 0x00, 0x00,
|
||||
0x20, 0x00, 0x04, 0x00, 0x1c, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
|
||||
0x06, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x03, 0x00, 0x1f, 0x00, 0x00, 0x00,
|
||||
0x06, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x03, 0x00, 0x20, 0x00, 0x00, 0x00,
|
||||
0x1f, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x21, 0x00, 0x00, 0x00,
|
||||
0x02, 0x00, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00,
|
||||
0x21, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
|
||||
0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x28, 0x00, 0x00, 0x00,
|
||||
0x1b, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x1c, 0x00, 0x00, 0x00,
|
||||
0x1d, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x03, 0x00,
|
||||
0x1e, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x03, 0x00,
|
||||
0x1f, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00,
|
||||
0x20, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x1f, 0x00, 0x00, 0x00,
|
||||
0x3b, 0x00, 0x04, 0x00, 0x20, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00,
|
||||
0x02, 0x00, 0x00, 0x00, 0x32, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
|
||||
0x22, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x32, 0x00, 0x04, 0x00,
|
||||
0x06, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
|
||||
0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x24, 0x00, 0x00, 0x00,
|
||||
0x01, 0x00, 0x00, 0x00, 0x2c, 0x00, 0x06, 0x00, 0x09, 0x00, 0x00, 0x00,
|
||||
0x29, 0x00, 0x00, 0x00, 0x28, 0x00, 0x00, 0x00, 0x28, 0x00, 0x00, 0x00,
|
||||
0x28, 0x00, 0x00, 0x00, 0x36, 0x00, 0x05, 0x00, 0x02, 0x00, 0x00, 0x00,
|
||||
0x25, 0x00, 0x00, 0x00, 0x24, 0x00, 0x00, 0x00, 0x24, 0x00, 0x00, 0x00,
|
||||
0x24, 0x00, 0x00, 0x00, 0x36, 0x00, 0x05, 0x00, 0x02, 0x00, 0x00, 0x00,
|
||||
0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00,
|
||||
0xf8, 0x00, 0x02, 0x00, 0x05, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00,
|
||||
0x07, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00,
|
||||
|
|
@ -118,23 +130,12 @@ static unsigned const char shaders_glsl_opmult_comp_spv[] = {
|
|||
0x06, 0x00, 0x00, 0x00, 0x0f, 0x00, 0x00, 0x00, 0x0e, 0x00, 0x00, 0x00,
|
||||
0x3e, 0x00, 0x03, 0x00, 0x08, 0x00, 0x00, 0x00, 0x0f, 0x00, 0x00, 0x00,
|
||||
0x3d, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x16, 0x00, 0x00, 0x00,
|
||||
0x08, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
|
||||
0x1b, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x41, 0x00, 0x06, 0x00,
|
||||
0x1c, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x00, 0x00, 0x1a, 0x00, 0x00, 0x00,
|
||||
0x15, 0x00, 0x00, 0x00, 0x1b, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00,
|
||||
0x06, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x00, 0x00,
|
||||
0x3d, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00,
|
||||
0x08, 0x00, 0x00, 0x00, 0x41, 0x00, 0x06, 0x00, 0x1c, 0x00, 0x00, 0x00,
|
||||
0x24, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0x15, 0x00, 0x00, 0x00,
|
||||
0x23, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
|
||||
0x25, 0x00, 0x00, 0x00, 0x24, 0x00, 0x00, 0x00, 0x84, 0x00, 0x05, 0x00,
|
||||
0x06, 0x00, 0x00, 0x00, 0x26, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x00, 0x00,
|
||||
0x25, 0x00, 0x00, 0x00, 0x41, 0x00, 0x06, 0x00, 0x1c, 0x00, 0x00, 0x00,
|
||||
0x27, 0x00, 0x00, 0x00, 0x13, 0x00, 0x00, 0x00, 0x15, 0x00, 0x00, 0x00,
|
||||
0x16, 0x00, 0x00, 0x00, 0x3e, 0x00, 0x03, 0x00, 0x27, 0x00, 0x00, 0x00,
|
||||
0x26, 0x00, 0x00, 0x00, 0xfd, 0x00, 0x01, 0x00, 0x38, 0x00, 0x01, 0x00
|
||||
0x08, 0x00, 0x00, 0x00, 0x41, 0x00, 0x06, 0x00, 0x18, 0x00, 0x00, 0x00,
|
||||
0x19, 0x00, 0x00, 0x00, 0x13, 0x00, 0x00, 0x00, 0x15, 0x00, 0x00, 0x00,
|
||||
0x16, 0x00, 0x00, 0x00, 0x3e, 0x00, 0x03, 0x00, 0x19, 0x00, 0x00, 0x00,
|
||||
0x17, 0x00, 0x00, 0x00, 0xfd, 0x00, 0x01, 0x00, 0x38, 0x00, 0x01, 0x00
|
||||
};
|
||||
static unsigned const int shaders_glsl_opmult_comp_spv_len = 1308;
|
||||
static unsigned const int shaders_glsl_opmult_comp_spv_len = 1320;
|
||||
}
|
||||
}
|
||||
#endif // define SHADEROP_SHADEROPMULT_HPP
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue