From b333b06772c89d96aacb5490d6a219fba7c09cc6 Mon Sep 17 00:00:00 2001 From: Mitja Felicijan Date: Thu, 12 Feb 2026 20:57:17 +0100 Subject: Engage! --- .../ggml-vulkan/vulkan-shaders/dequant_q4_1.comp | 32 ++++++++++++++++++++++ 1 file changed, 32 insertions(+) create mode 100644 llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q4_1.comp (limited to 'llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q4_1.comp') diff --git a/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q4_1.comp b/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q4_1.comp new file mode 100644 index 0000000..6b63cbe --- /dev/null +++ b/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q4_1.comp @@ -0,0 +1,32 @@ +#version 450 + +#include "dequant_head.glsl" + +layout(local_size_x = 256, local_size_y = 1, local_size_z = 1) in; + +layout (binding = 0) readonly buffer A {block_q4_1 data_a[];}; +layout (binding = 1) writeonly buffer D {D_TYPE data_b[];}; + +void main() { + const uint i = gl_WorkGroupID.x * 4 + gl_LocalInvocationID.x / 64; + + const uint tid = gl_LocalInvocationID.x % 64; + const uint il = tid/32; + const uint ir = tid%32; + const uint ib = 32*i + ir; + if (ib >= p.nel / 32) { + return; + } + + const uint b_idx = 1024*i + 32*ir + 8*il; + + const float d = float(data_a[ib].d); + const float m = float(data_a[ib].m); + + const uint q_idx = 8*il; + + [[unroll]] for (uint l = 0; l < 8; ++l) { + data_b[b_idx + l + 0] = D_TYPE(d * (data_a[ib].qs[q_idx + l] & 0xF) + m); + data_b[b_idx + l + 16] = D_TYPE(d * (data_a[ib].qs[q_idx + l] >> 4) + m); + } +} -- cgit v1.2.3