Skip to content

Commit e1575ed

Browse files
stduhpfstduhpf
committed
Vulkan: RTE rounding for cpy to quant
1 parent e046430 commit e1575ed

File tree

3 files changed

+46
-17
lines changed

3 files changed

+46
-17
lines changed

‎ggml/src/ggml-vulkan/ggml-vulkan.cpp‎

Lines changed: 40 additions & 17 deletions
Original file line numberDiff line numberDiff line change
@@ -299,6 +299,7 @@ struct vk_device_struct {
299299
vk_pipeline pipeline_cpy_f32_f32, pipeline_cpy_f32_f16, pipeline_cpy_f16_f16;
300300
vk_pipeline pipeline_contig_cpy_f32_f32, pipeline_contig_cpy_f32_f16, pipeline_contig_cpy_f16_f16;
301301
vk_pipeline pipeline_cpy_f32_quant[GGML_TYPE_COUNT];
302+
vk_pipeline pipeline_cpy_f32_quant_rte[GGML_TYPE_COUNT];
302303
vk_pipeline pipeline_cpy_quant_f32[GGML_TYPE_COUNT];
303304
vk_pipeline pipeline_norm_f32;
304305
vk_pipeline pipeline_group_norm_f32;
@@ -2281,13 +2282,21 @@ static void ggml_vk_load_shaders(vk_device& device) {
22812282
ggml_vk_create_pipeline(device, device->pipeline_contig_cpy_f32_f32, "contig_cpy_f32_f32", contig_cpy_f32_f32_len, contig_cpy_f32_f32_data, "main", 2, sizeof(vk_op_unary_push_constants), {512, 1, 1}, {}, 1);
22822283
ggml_vk_create_pipeline(device, device->pipeline_contig_cpy_f32_f16, "contig_cpy_f32_f16", contig_cpy_f32_f16_len, contig_cpy_f32_f16_data, "main", 2, sizeof(vk_op_unary_push_constants), {512, 1, 1}, {}, 1);
22832284
ggml_vk_create_pipeline(device, device->pipeline_contig_cpy_f16_f16, "contig_cpy_f16_f16", contig_cpy_f16_f16_len, contig_cpy_f16_f16_data, "main", 2, sizeof(vk_op_unary_push_constants), {512, 1, 1}, {}, 1);
2284-
2285-
ggml_vk_create_pipeline(device, device->pipeline_cpy_f32_quant[GGML_TYPE_Q4_0], "cpy_f32_q4_0", cpy_f32_q4_0_len, cpy_f32_q4_0_data, "main", 2, sizeof(vk_op_unary_push_constants), {(uint32_t)ggml_blck_size(GGML_TYPE_Q4_0), 1, 1}, {}, 1);
2286-
ggml_vk_create_pipeline(device, device->pipeline_cpy_f32_quant[GGML_TYPE_Q4_1], "cpy_f32_q4_1", cpy_f32_q4_1_len, cpy_f32_q4_1_data, "main", 2, sizeof(vk_op_unary_push_constants), {(uint32_t)ggml_blck_size(GGML_TYPE_Q4_1), 1, 1}, {}, 1);
2287-
ggml_vk_create_pipeline(device, device->pipeline_cpy_f32_quant[GGML_TYPE_Q5_0], "cpy_f32_q5_0", cpy_f32_q5_0_len, cpy_f32_q5_0_data, "main", 2, sizeof(vk_op_unary_push_constants), {(uint32_t)ggml_blck_size(GGML_TYPE_Q5_0), 1, 1}, {}, 1);
2288-
ggml_vk_create_pipeline(device, device->pipeline_cpy_f32_quant[GGML_TYPE_Q5_1], "cpy_f32_q5_1", cpy_f32_q5_1_len, cpy_f32_q5_1_data, "main", 2, sizeof(vk_op_unary_push_constants), {(uint32_t)ggml_blck_size(GGML_TYPE_Q5_1), 1, 1}, {}, 1);
2289-
ggml_vk_create_pipeline(device, device->pipeline_cpy_f32_quant[GGML_TYPE_Q8_0], "cpy_f32_q8_0", cpy_f32_q8_0_len, cpy_f32_q8_0_data, "main", 2, sizeof(vk_op_unary_push_constants), {(uint32_t)ggml_blck_size(GGML_TYPE_Q8_0), 1, 1}, {}, 1);
2290-
ggml_vk_create_pipeline(device, device->pipeline_cpy_f32_quant[GGML_TYPE_IQ4_NL], "cpy_f32_iq4_nl", cpy_f32_iq4_nl_len, cpy_f32_iq4_nl_data, "main", 2, sizeof(vk_op_unary_push_constants), {(uint32_t)ggml_blck_size(GGML_TYPE_IQ4_NL), 1, 1}, {}, 1);
2285+
if (device->float_controls_rte_fp16) {
2286+
ggml_vk_create_pipeline(device, device->pipeline_cpy_f32_quant_rte[GGML_TYPE_Q4_0], "cpy_f32_q4_0", cpy_f32_q4_0_len, cpy_f32_q4_0_data, "main", 2, sizeof(vk_op_unary_push_constants), {(uint32_t)ggml_blck_size(GGML_TYPE_Q4_0), 1, 1}, {}, 1);
2287+
ggml_vk_create_pipeline(device, device->pipeline_cpy_f32_quant_rte[GGML_TYPE_Q4_1], "cpy_f32_q4_1", cpy_f32_q4_1_rte_len, cpy_f32_q4_1_rte_data, "main", 2, sizeof(vk_op_unary_push_constants), {(uint32_t)ggml_blck_size(GGML_TYPE_Q4_1), 1, 1}, {}, 1);
2288+
ggml_vk_create_pipeline(device, device->pipeline_cpy_f32_quant_rte[GGML_TYPE_Q5_0], "cpy_f32_q5_0", cpy_f32_q5_0_rte_len, cpy_f32_q5_0_rte_data, "main", 2, sizeof(vk_op_unary_push_constants), {(uint32_t)ggml_blck_size(GGML_TYPE_Q5_0), 1, 1}, {}, 1);
2289+
ggml_vk_create_pipeline(device, device->pipeline_cpy_f32_quant_rte[GGML_TYPE_Q5_1], "cpy_f32_q5_1", cpy_f32_q5_1_rte_len, cpy_f32_q5_1_rte_data, "main", 2, sizeof(vk_op_unary_push_constants), {(uint32_t)ggml_blck_size(GGML_TYPE_Q5_1), 1, 1}, {}, 1);
2290+
ggml_vk_create_pipeline(device, device->pipeline_cpy_f32_quant_rte[GGML_TYPE_Q8_0], "cpy_f32_q8_0", cpy_f32_q8_0_rte_len, cpy_f32_q8_0_rte_data, "main", 2, sizeof(vk_op_unary_push_constants), {(uint32_t)ggml_blck_size(GGML_TYPE_Q8_0), 1, 1}, {}, 1);
2291+
ggml_vk_create_pipeline(device, device->pipeline_cpy_f32_quant_rte[GGML_TYPE_IQ4_NL], "cpy_f32_iq4_nl", cpy_f32_iq4_nl_rte_len, cpy_f32_iq4_nl_rte_data, "main", 2, sizeof(vk_op_unary_push_constants), {(uint32_t)ggml_blck_size(GGML_TYPE_IQ4_NL), 1, 1}, {}, 1);
2292+
} else {
2293+
ggml_vk_create_pipeline(device, device->pipeline_cpy_f32_quant[GGML_TYPE_Q4_0], "cpy_f32_q4_0", cpy_f32_q4_0_len, cpy_f32_q4_0_data, "main", 2, sizeof(vk_op_unary_push_constants), {(uint32_t)ggml_blck_size(GGML_TYPE_Q4_0), 1, 1}, {}, 1);
2294+
ggml_vk_create_pipeline(device, device->pipeline_cpy_f32_quant[GGML_TYPE_Q4_1], "cpy_f32_q4_1", cpy_f32_q4_1_len, cpy_f32_q4_1_data, "main", 2, sizeof(vk_op_unary_push_constants), {(uint32_t)ggml_blck_size(GGML_TYPE_Q4_1), 1, 1}, {}, 1);
2295+
ggml_vk_create_pipeline(device, device->pipeline_cpy_f32_quant[GGML_TYPE_Q5_0], "cpy_f32_q5_0", cpy_f32_q5_0_len, cpy_f32_q5_0_data, "main", 2, sizeof(vk_op_unary_push_constants), {(uint32_t)ggml_blck_size(GGML_TYPE_Q5_0), 1, 1}, {}, 1);
2296+
ggml_vk_create_pipeline(device, device->pipeline_cpy_f32_quant[GGML_TYPE_Q5_1], "cpy_f32_q5_1", cpy_f32_q5_1_len, cpy_f32_q5_1_data, "main", 2, sizeof(vk_op_unary_push_constants), {(uint32_t)ggml_blck_size(GGML_TYPE_Q5_1), 1, 1}, {}, 1);
2297+
ggml_vk_create_pipeline(device, device->pipeline_cpy_f32_quant[GGML_TYPE_Q8_0], "cpy_f32_q8_0", cpy_f32_q8_0_len, cpy_f32_q8_0_data, "main", 2, sizeof(vk_op_unary_push_constants), {(uint32_t)ggml_blck_size(GGML_TYPE_Q8_0), 1, 1}, {}, 1);
2298+
ggml_vk_create_pipeline(device, device->pipeline_cpy_f32_quant[GGML_TYPE_IQ4_NL], "cpy_f32_iq4_nl", cpy_f32_iq4_nl_len, cpy_f32_iq4_nl_data, "main", 2, sizeof(vk_op_unary_push_constants), {(uint32_t)ggml_blck_size(GGML_TYPE_IQ4_NL), 1, 1}, {}, 1);
2299+
}
22912300

22922301
ggml_vk_create_pipeline(device, device->pipeline_cpy_quant_f32[GGML_TYPE_Q4_0], "cpy_q4_0_f32", cpy_q4_0_f32_len, cpy_q4_0_f32_data, "main", 2, sizeof(vk_op_unary_push_constants), {(uint32_t)ggml_blck_size(GGML_TYPE_Q4_0), 1, 1}, {}, 1);
22932302
ggml_vk_create_pipeline(device, device->pipeline_cpy_quant_f32[GGML_TYPE_Q4_1], "cpy_q4_1_f32", cpy_q4_1_f32_len, cpy_q4_1_f32_data, "main", 2, sizeof(vk_op_unary_push_constants), {(uint32_t)ggml_blck_size(GGML_TYPE_Q4_1), 1, 1}, {}, 1);
@@ -4133,16 +4142,30 @@ static vk_pipeline ggml_vk_get_cpy_pipeline(ggml_backend_vk_context * ctx, const
41334142
}
41344143
}
41354144
if (src->type == GGML_TYPE_F32) {
4136-
switch (to) {
4137-
case GGML_TYPE_Q4_0:
4138-
case GGML_TYPE_Q4_1:
4139-
case GGML_TYPE_Q5_0:
4140-
case GGML_TYPE_Q5_1:
4141-
case GGML_TYPE_Q8_0:
4142-
case GGML_TYPE_IQ4_NL:
4143-
return ctx->device->pipeline_cpy_f32_quant[to];
4144-
default:
4145-
break;
4145+
if (ctx->device->float_controls_rte_fp16) {
4146+
switch (to) {
4147+
case GGML_TYPE_Q4_0:
4148+
case GGML_TYPE_Q4_1:
4149+
case GGML_TYPE_Q5_0:
4150+
case GGML_TYPE_Q5_1:
4151+
case GGML_TYPE_Q8_0:
4152+
case GGML_TYPE_IQ4_NL:
4153+
return ctx->device->pipeline_cpy_f32_quant_rte[to];
4154+
default:
4155+
break;
4156+
}
4157+
} else {
4158+
switch (to) {
4159+
case GGML_TYPE_Q4_0:
4160+
case GGML_TYPE_Q4_1:
4161+
case GGML_TYPE_Q5_0:
4162+
case GGML_TYPE_Q5_1:
4163+
case GGML_TYPE_Q8_0:
4164+
case GGML_TYPE_IQ4_NL:
4165+
return ctx->device->pipeline_cpy_f32_quant[to];
4166+
default:
4167+
break;
4168+
}
41464169
}
41474170
}
41484171

‎ggml/src/ggml-vulkan/vulkan-shaders/copy_to_quant.comp‎

Lines changed: 5 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,10 @@
11
#version 450
22

3+
#if RTE16
4+
#extension GL_EXT_spirv_intrinsics : enable
5+
spirv_execution_mode(capabilities = [4467], 4462, 16); // RoundingModeRTE, 16 bits
6+
#endif // RTE16
7+
38
#include "types.comp"
49
#include "generic_unary_head.comp"
510

‎ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -445,6 +445,7 @@ void process_shaders() {
445445

446446
for (std::string t : {"q4_0", "q4_1", "q5_0", "q5_1", "q8_0", "iq4_nl"}) {
447447
string_to_spv("cpy_f32_" + t, "copy_to_quant.comp", {{"DATA_A_" + to_uppercase(t), "1"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}});
448+
string_to_spv("cpy_f32_" + t + "_rte", "copy_to_quant.comp", {{"DATA_A_" + to_uppercase(t), "1"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}, {"RTE16", "1"}});
448449
string_to_spv("cpy_" + t + "_f32", "copy_from_quant.comp", {{"DATA_A_" + to_uppercase(t), "1"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}});
449450
}
450451

0 commit comments

Comments
 (0)