From 9e40d984ad27d7b60392fb2b7548885201864fe4 Mon Sep 17 00:00:00 2001
From: Aaron Teo <aaron.teo1@ibm.com>
Date: Wed, 25 Jun 2025 14:57:58 +0800
Subject: [PATCH] ggml: move ggml_table_f32_f16 to ggml-cpu

ref: https://github.com/ggml-org/llama.cpp/pull/14317#discussion_r2164775006

Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
---
 ggml/src/ggml-cpu/ggml-cpu.c      |  1 +
 ggml/src/ggml-cpu/simd-mappings.h |  4 ++++
 ggml/src/ggml-impl.h              |  4 ----
 ggml/src/ggml.c                   | 21 ---------------------
 4 files changed, 5 insertions(+), 25 deletions(-)

diff --git a/ggml/src/ggml-cpu/ggml-cpu.c b/ggml/src/ggml-cpu/ggml-cpu.c
index 70f328015..ce296898c 100644
--- a/ggml/src/ggml-cpu/ggml-cpu.c
+++ b/ggml/src/ggml-cpu/ggml-cpu.c
@@ -3479,6 +3479,7 @@ void ggml_cpu_init(void) {
                     ggml_fp16_t fp16;
                 } u = {i};
                 float f = GGML_CPU_FP16_TO_FP32(u.fp16);
+                ggml_table_f32_f16[i] = GGML_COMPUTE_FP16_TO_FP32(u.fp16);
                 ggml_table_gelu_f16[i] = GGML_CPU_FP32_TO_FP16(ggml_gelu_f32(f));
                 ggml_table_gelu_quick_f16[i] = GGML_CPU_FP32_TO_FP16(ggml_gelu_quick_f32(f));
             }
diff --git a/ggml/src/ggml-cpu/simd-mappings.h b/ggml/src/ggml-cpu/simd-mappings.h
index 655ab3c6c..2f65ccd17 100644
--- a/ggml/src/ggml-cpu/simd-mappings.h
+++ b/ggml/src/ggml-cpu/simd-mappings.h
@@ -137,6 +137,10 @@
     }
 #endif
 
+// precomputed f32 table for f16 (256 KB)
+// defined in ggml.c, initialized in ggml_init()
+GGML_API float ggml_table_f32_f16[1 << 16];
+
 // On ARM NEON, it's quicker to directly convert x -> x instead of calling into ggml_lookup_fp16_to_fp32,
 // so we define GGML_CPU_FP16_TO_FP32 and GGML_CPU_FP32_TO_FP16 elsewhere for NEON.
 // This is also true for POWER9.
diff --git a/ggml/src/ggml-impl.h b/ggml/src/ggml-impl.h
index 8d9bdc74c..57761644f 100644
--- a/ggml/src/ggml-impl.h
+++ b/ggml/src/ggml-impl.h
@@ -393,10 +393,6 @@ static inline ggml_fp16_t ggml_compute_fp32_to_fp16(float f) {
 #define GGML_FP16_TO_FP32(x) GGML_COMPUTE_FP16_TO_FP32(x)
 #define GGML_FP32_TO_FP16(x) GGML_COMPUTE_FP32_TO_FP16(x)
 
-// precomputed f32 table for f16 (256 KB)
-// defined in ggml.c, initialized in ggml_init()
-GGML_API float ggml_table_f32_f16[1 << 16];
-
 /**
  * Converts brain16 to float32.
  *
diff --git a/ggml/src/ggml.c b/ggml/src/ggml.c
index f8e7c595b..e0e462882 100644
--- a/ggml/src/ggml.c
+++ b/ggml/src/ggml.c
@@ -1414,27 +1414,6 @@ static inline bool ggml_can_repeat_rows(const struct ggml_tensor * t0, const str
 ////////////////////////////////////////////////////////////////////////////////
 
 struct ggml_context * ggml_init(struct ggml_init_params params) {
-    static bool is_first_call = true;
-
-    ggml_critical_section_start();
-
-    if (is_first_call) {
-        // initialize time system (required on Windows)
-        ggml_time_init();
-
-        for (int i = 0; i < (1 << 16); ++i) {
-            union {
-                uint16_t u16;
-                ggml_fp16_t fp16;
-            } u = {i};
-            ggml_table_f32_f16[i] = GGML_COMPUTE_FP16_TO_FP32(u.fp16);
-        }
-
-        is_first_call = false;
-    }
-
-    ggml_critical_section_end();
-
     struct ggml_context * ctx = GGML_MALLOC(sizeof(struct ggml_context));
 
     // allow to call ggml_init with 0 size