summaryrefslogtreecommitdiff
path: root/vulkan-shaders/dequant_q5_1.comp
diff options
context:
space:
mode:
Diffstat (limited to 'vulkan-shaders/dequant_q5_1.comp')
-rw-r--r--vulkan-shaders/dequant_q5_1.comp35
1 files changed, 35 insertions, 0 deletions
diff --git a/vulkan-shaders/dequant_q5_1.comp b/vulkan-shaders/dequant_q5_1.comp
new file mode 100644
index 00000000..dc59fe3b
--- /dev/null
+++ b/vulkan-shaders/dequant_q5_1.comp
@@ -0,0 +1,35 @@
+#version 450
+
+#include "dequant_head.comp"
+
+layout(local_size_x = 256, local_size_y = 1, local_size_z = 1) in;
+
+layout (binding = 0) readonly buffer A {block_q5_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 qh = data_a[ib].qh;
+
+ const uint q_idx = 8*il;
+
+ [[unroll]] for (uint l = 0; l < 8; ++l) {
+ const uint iqs = q_idx + l;
+ const uint vui = uint(data_a[ib].qs[iqs]);
+ data_b[b_idx + l + 0] = D_TYPE(d * (((vui & 0xF) | (((qh >> iqs) << 4) & 0x10))) + m);
+ data_b[b_idx + l + 16] = D_TYPE(d * (((vui >> 4) | ((qh >> (iqs + 12)) & 0x10))) + m);
+ }
+}