Skip to content

Commit cbce430

Browse files
vkhaitanvkhaitan
committed
Revert "vulkan: Support Q1_0 (ggml-org#21539)"
This reverts commit 7b69125.
1 parent 190ab77 commit cbce430

9 files changed

Lines changed: 4 additions & 161 deletions

File tree

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

Lines changed: 0 additions & 33 deletions
Large diffs are not rendered by default.

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

Lines changed: 0 additions & 25 deletions
Original file line numberDiff line numberDiff line change
@@ -184,31 +184,6 @@ void quantize(uint dst_idx, uint src_idx)
184184
}
185185
#endif
186186

187-
#if defined(DATA_A_Q1_0)
188-
void quantize(uint dst_idx, uint src_idx)
189-
{
190-
float sum_abs = 0.0;
191-
192-
[[unroll]] for (int j = 0; j < QUANT_K_Q1_0; j++) {
193-
sum_abs += abs(data_s[src_idx + j]);
194-
}
195-
196-
const float d = sum_abs / QUANT_K_Q1_0;
197-
198-
data_q[dst_idx].d = float16_t(d);
199-
200-
[[unroll]] for (int j = 0; j < QUANT_K_Q1_0 / 8; ++j) {
201-
data_q[dst_idx].qs[j] = uint8_t(0);
202-
}
203-
204-
[[unroll]] for (int j = 0; j < QUANT_K_Q1_0; ++j) {
205-
if (data_s[src_idx + j] >= 0.0) {
206-
data_q[dst_idx].qs[j / 8] |= uint8_t(1 << (j % 8));
207-
}
208-
}
209-
}
210-
#endif
211-
212187
#if defined(DATA_A_IQ4_NL)
213188
uint best_index(float x) {
214189
if (x <= kvalues_iq4nl[0]) return 0;

‎ggml/src/ggml-vulkan/vulkan-shaders/dequant_funcs.glsl‎

Lines changed: 0 additions & 24 deletions
Original file line numberDiff line numberDiff line change
@@ -87,23 +87,6 @@ vec4 dequantize4(uint ib, uint iqs, uint a_offset) {
8787
}
8888
#endif
8989

90-
#if defined(DATA_A_Q1_0)
91-
vec2 dequantize(uint ib, uint iqs, uint a_offset) {
92-
const uint bits = uint(data_a[a_offset + ib].qs[iqs / 8u]) >> (iqs % 8u);
93-
return vec2(
94-
(bits & 1u) != 0u ? 1.0f : -1.0f,
95-
(bits & 2u) != 0u ? 1.0f : -1.0f);
96-
}
97-
vec4 dequantize4(uint ib, uint iqs, uint a_offset) {
98-
const uint bits = uint(data_a[a_offset + ib].qs[iqs / 8u]) >> (iqs % 8u);
99-
return vec4(
100-
(bits & 1u) != 0u ? 1.0f : -1.0f,
101-
(bits & 2u) != 0u ? 1.0f : -1.0f,
102-
(bits & 4u) != 0u ? 1.0f : -1.0f,
103-
(bits & 8u) != 0u ? 1.0f : -1.0f);
104-
}
105-
#endif
106-
10790
#if defined(DATA_A_IQ1_S)
10891
vec2 dequantize(uint ib, uint iqs, uint a_offset) {
10992
const uint ib32 = iqs / 32;
@@ -471,13 +454,6 @@ vec2 get_dm(uint ib, uint a_offset) {
471454
}
472455
#endif
473456

474-
#if defined(DATA_A_Q1_0)
475-
vec2 get_dm(uint ib, uint a_offset) {
476-
const float d = float(data_a[a_offset + ib].d);
477-
return vec2(d, 0);
478-
}
479-
#endif
480-
481457
#if defined(DATA_A_MXFP4)
482458
vec2 get_dm(uint ib, uint a_offset) {
483459
return vec2(e8m0_to_fp32(data_a[a_offset + ib].e), 0);

‎ggml/src/ggml-vulkan/vulkan-shaders/dequant_funcs_cm2.glsl‎

Lines changed: 1 addition & 15 deletions
Original file line numberDiff line numberDiff line change
@@ -13,18 +13,6 @@ float16_t dequantFuncF32(const in decodeBufF32 bl, const in uint blockCoords[2],
1313
return vf16[idx];
1414
}
1515

16-
layout(buffer_reference, std430, buffer_reference_align = 2) buffer decodeBufQ1_0 {
17-
block_q1_0 block;
18-
};
19-
20-
float16_t dequantFuncQ1_0(const in decodeBufQ1_0 bl, const in uint blockCoords[2], const in uint coordInBlock[2])
21-
{
22-
const float16_t d = bl.block.d;
23-
const uint idx = coordInBlock[1];
24-
const uint bit = (uint(bl.block.qs[(idx & 0x78) >> 3]) >> (idx & 0x7)) & 1u;
25-
return bit != 0u ? d : -d;
26-
}
27-
2816
layout(buffer_reference, std430, buffer_reference_align = 2) buffer decodeBufQ4_0 {
2917
block_q4_0_packed16 block;
3018
};
@@ -697,9 +685,7 @@ float16_t dequantFuncMXFP4(const in decodeBufMXFP4 bl, const in uint blockCoords
697685
}
698686
#endif
699687

700-
#if defined(DATA_A_Q1_0)
701-
#define dequantFuncA dequantFuncQ1_0
702-
#elif defined(DATA_A_Q4_0)
688+
#if defined(DATA_A_Q4_0)
703689
#define dequantFuncA dequantFuncQ4_0
704690
#elif defined(DATA_A_Q4_1)
705691
#define dequantFuncA dequantFuncQ4_1

‎ggml/src/ggml-vulkan/vulkan-shaders/dequant_q1_0.comp‎

Lines changed: 0 additions & 29 deletions
This file was deleted.

‎ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_funcs.glsl‎

Lines changed: 0 additions & 14 deletions
Original file line numberDiff line numberDiff line change
@@ -130,20 +130,6 @@ void load_a_to_shmem(const uint pos_a, const uint row, const uint col, const uin
130130

131131
buf_a[buf_idx ] = FLOAT_TYPEV2(v.xy);
132132
buf_a[buf_idx + 1] = FLOAT_TYPEV2(v.zw);
133-
#elif defined(DATA_A_Q1_0)
134-
const uint idx = pos_a + col * p.stride_a / LOAD_VEC_A + row;
135-
const uint buf_idx = col * SHMEM_STRIDE + row * LOAD_VEC_A / 2;
136-
137-
const uint ib = idx / 16;
138-
const uint iqs = idx & 0xfu;
139-
140-
const float d = float(data_a[ib].d);
141-
const uint bits = uint(data_a[ib].qs[iqs]);
142-
143-
buf_a[buf_idx ] = FLOAT_TYPEV2((bits & 0x01u) != 0u ? d : -d, (bits & 0x02u) != 0u ? d : -d);
144-
buf_a[buf_idx + 1] = FLOAT_TYPEV2((bits & 0x04u) != 0u ? d : -d, (bits & 0x08u) != 0u ? d : -d);
145-
buf_a[buf_idx + 2] = FLOAT_TYPEV2((bits & 0x10u) != 0u ? d : -d, (bits & 0x20u) != 0u ? d : -d);
146-
buf_a[buf_idx + 3] = FLOAT_TYPEV2((bits & 0x40u) != 0u ? d : -d, (bits & 0x80u) != 0u ? d : -d);
147133
#elif defined(DATA_A_Q2_K)
148134
const uint idx = pos_a + col * p.stride_a / LOAD_VEC_A + row;
149135
const uint buf_idx = col * SHMEM_STRIDE + row * LOAD_VEC_A / 2;

‎ggml/src/ggml-vulkan/vulkan-shaders/types.glsl‎

Lines changed: 0 additions & 16 deletions
Original file line numberDiff line numberDiff line change
@@ -188,22 +188,6 @@ struct block_q8_0_packed16
188188
#define DATA_A_QUANT_LEGACY
189189
#endif
190190

191-
#define QUANT_K_Q1_0 128
192-
#define QUANT_R_Q1_0 1
193-
194-
struct block_q1_0
195-
{
196-
float16_t d;
197-
uint8_t qs[QUANT_K_Q1_0 / 8];
198-
};
199-
200-
#if defined(DATA_A_Q1_0)
201-
#define QUANT_K QUANT_K_Q1_0
202-
#define QUANT_R QUANT_R_Q1_0
203-
#define QUANT_AUXF 1
204-
#define A_TYPE block_q1_0
205-
#endif
206-
207191
#define QUANT_K_Q8_1 32
208192
#define QUANT_R_Q8_1 1
209193

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

Lines changed: 3 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -45,7 +45,6 @@ std::string target_cpp = "";
4545
const std::vector<std::string> type_names = {
4646
"f32",
4747
"f16",
48-
"q1_0",
4948
"q4_0",
5049
"q4_1",
5150
"q5_0",
@@ -554,7 +553,7 @@ void matmul_shaders(bool fp16, MatMulIdType matmul_id_type, bool coopmat, bool c
554553

555554
for (const auto& tname : type_names) {
556555
std::string load_vec_quant = "2";
557-
if ((tname == "q1_0") || (tname == "q4_0") || (tname == "q4_1") || (tname == "q5_1") || (tname == "iq1_s") || (tname == "iq1_m") || (tname == "iq2_xxs") || (tname == "iq2_xs") || (tname == "iq2_s"))
556+
if ((tname == "q4_0") || (tname == "q4_1") || (tname == "q5_1") || (tname == "iq1_s") || (tname == "iq1_m") || (tname == "iq2_xxs") || (tname == "iq2_xs") || (tname == "iq2_s"))
558557
load_vec_quant = "8";
559558
else if ((tname == "q5_0") || (tname == "q8_0") || (tname == "q2_k") || (tname == "q4_k") || (tname == "q5_k") || (tname == "iq3_xxs") || (tname == "iq3_s") || (tname == "iq4_xs") || (tname == "iq4_nl") || (tname == "mxfp4"))
560559
load_vec_quant = "4";
@@ -759,13 +758,13 @@ void process_shaders() {
759758
string_to_spv("cpy_transpose_16", "copy_transpose.comp", {{"A_TYPE", "uint16_t"}, {"D_TYPE", "uint16_t"}});
760759
string_to_spv("cpy_transpose_32", "copy_transpose.comp", {{"A_TYPE", "uint"}, {"D_TYPE", "uint"}});
761760

762-
for (std::string t : {"q1_0", "q4_0", "q4_1", "q5_0", "q5_1", "q8_0", "iq4_nl"}) {
761+
for (std::string t : {"q4_0", "q4_1", "q5_0", "q5_1", "q8_0", "iq4_nl"}) {
763762
string_to_spv("cpy_f32_" + t, "copy_to_quant.comp", {{"DATA_A_" + to_uppercase(t), "1"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}});
764763
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"}});
765764
string_to_spv("cpy_" + t + "_f32", "copy_from_quant.comp", {{"DATA_A_" + to_uppercase(t), "1"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}});
766765
}
767766

768-
for (std::string t : {"f32", "f16", "bf16", "q1_0", "q4_0", "q4_1", "q5_0", "q5_1", "q8_0", "iq4_nl"}) {
767+
for (std::string t : {"f32", "f16", "bf16", "q4_0", "q4_1", "q5_0", "q5_1", "q8_0", "iq4_nl"}) {
769768
string_to_spv("set_rows_" + t + "_i32", "copy_to_quant.comp", {{"SET_ROWS", "1"}, {"DATA_A_" + to_uppercase(t), "1"}, {"B_TYPE", "uint"}, {"B_SIZE", "32"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}});
770769
string_to_spv("set_rows_" + t + "_i32_rte", "copy_to_quant.comp", {{"SET_ROWS", "1"}, {"DATA_A_" + to_uppercase(t), "1"}, {"B_TYPE", "uint"}, {"B_SIZE", "32"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}, {"RTE16", "1"}});
771770
string_to_spv("set_rows_" + t + "_i64", "copy_to_quant.comp", {{"SET_ROWS", "1"}, {"DATA_A_" + to_uppercase(t), "1"}, {"B_TYPE", "uvec2"}, {"B_SIZE", "64"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}});

‎tests/test-backend-ops.cpp‎

Lines changed: 0 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -7265,7 +7265,6 @@ static const ggml_type all_types[] = {
72657265
static const ggml_type base_types[] = {
72667266
GGML_TYPE_F32, GGML_TYPE_F16,
72677267
GGML_TYPE_Q8_0, // for I8MM tests
7268-
GGML_TYPE_Q1_0,
72697268
GGML_TYPE_Q4_0,
72707269
GGML_TYPE_Q4_1, // for I8MM tests
72717270
GGML_TYPE_Q4_K,

0 commit comments

Comments
 (0)