From 08ffba21e5c2f1728b6b405c05ff3913f65706d4 Mon Sep 17 00:00:00 2001 From: Amadeus dyw <74057036+SG-Amadeus@users.noreply.github.com> Date: Mon, 12 Oct 2026 14:00:26 +0800 Subject: [PATCH] vulkan: use ternary expressions to avoid switch lowering for CM2 quant block size (#30164) * vulkan: specialize quant block size for CM2 matmul * vulkan: use ternary expressions to avoid switch lowering for CM2 quant block size * vulkan: clarify CM2 quant block size comment --- .../ggml-vulkan/vulkan-shaders/mul_mm_cm2.comp | 17 ++++++----------- 1 file changed, 6 insertions(+), 11 deletions(-) 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 e3f1dd5ab..89022f047 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