Skip to content

LLM inference in C/C++

C++ ggml

vulkan: use ternary expressions to avoid switch lowering for CM2 quant block size (#30164)

Amadeus dyw08ffba21e5c2
diff --git a/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_cm2.comp b/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_cm2.comp
index e3f1dd5abb..89022f0478 100644
--- a/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_cm2.comp
+++ b/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_cm2.comp
@@ -49,17 +49,12 @@ layout (constant_id = 6) const uint subgroup_size = 32;
#ifdef MULMAT_QUANT
uint mm_quant_k() {
- switch (MmTypeA) {
- case GGML_TYPE_Q4_0: case GGML_TYPE_Q4_1: case GGML_TYPE_Q5_0: case GGML_TYPE_Q5_1:
- case GGML_TYPE_Q8_0:
- return 32u;
- case GGML_TYPE_Q1_0:
- return 128u;
- case GGML_TYPE_Q2_0:
- return 64u;
- default:
- return 256u;
- }
+ // Use ternary expressions instead of a switch to keep qk as a specialization-constant expression.
+ return (MmTypeA == GGML_TYPE_Q1_0) ? 128u :
+ (MmTypeA == GGML_TYPE_Q2_0) ? 64u :
+ ((MmTypeA == GGML_TYPE_Q4_0 || MmTypeA == GGML_TYPE_Q4_1 ||
+ MmTypeA == GGML_TYPE_Q5_0 || MmTypeA == GGML_TYPE_Q5_1 ||
+ MmTypeA == GGML_TYPE_Q8_0) ? 32u : 256u);
}
#endif