vulkan: add RTE variants for glu/add/sub/mul/div (#14653)

This commit is contained in:
Jeff Bolz 2025-07-15 14:32:11 -05:00 committed by GitHub
parent 68e37a61a7
commit 10a0351a97
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
8 changed files with 90 additions and 32 deletions

View file

@ -2835,10 +2835,11 @@ static void ggml_vk_load_shaders(vk_device& device) {
return s;
};
bool rte = device->float_controls_rte_fp16;
#define CREATE_BINARY(name, namemod, spec) \
for (int s0 : {0,1}) for (int s1 : {0,1}) for (int d : {0,1}) \
ggml_vk_create_pipeline(device, device->pipeline_ ## name ## namemod[s0][s1][d], \
#name + get_suffix(s0, s1, d) + #namemod, name ## _len[s0][s1][d], name ## _data[s0][s1][d], \
#name + get_suffix(s0, s1, d) + #namemod, name ## _len[s0][s1][d][rte], name ## _data[s0][s1][d][rte], \
"main", 3, sizeof(vk_op_binary_push_constants), {512, 1, 1}, spec, 1);
CREATE_BINARY(add, , {0})
@ -2890,8 +2891,13 @@ static void ggml_vk_load_shaders(vk_device& device) {
#undef CREATE_UNARY
#define CREATE_GLU(name) \
ggml_vk_create_pipeline(device, device->pipeline_ ## name [0], #name "_f32", name ## _f32_len, name ## _f32_data, "main", 3, sizeof(vk_op_glu_push_constants), {512, 1, 1}, {}, 1, true); \
ggml_vk_create_pipeline(device, device->pipeline_ ## name [1], #name "_f16", name ## _f16_len, name ## _f16_data, "main", 3, sizeof(vk_op_glu_push_constants), {512, 1, 1}, {}, 1, true);
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); \
ggml_vk_create_pipeline(device, device->pipeline_ ## name [1], #name "_f16_rte", name ## _f16_rte_len, name ## _f16_rte_data, "main", 3, sizeof(vk_op_glu_push_constants), {512, 1, 1}, {}, 1, true); \
} else { \
ggml_vk_create_pipeline(device, device->pipeline_ ## name [0], #name "_f32", name ## _f32_len, name ## _f32_data, "main", 3, sizeof(vk_op_glu_push_constants), {512, 1, 1}, {}, 1, true); \
ggml_vk_create_pipeline(device, device->pipeline_ ## name [1], #name "_f16", name ## _f16_len, name ## _f16_data, "main", 3, sizeof(vk_op_glu_push_constants), {512, 1, 1}, {}, 1, true); \
}
CREATE_GLU(geglu)
CREATE_GLU(reglu)