From 5bcd57722f2dbfc2f1e09447e11e3dee9c4b7c1c Mon Sep 17 00:00:00 2001
From: Reese Levine <reeselevine1@gmail.com>
Date: Fri, 14 Nov 2025 09:53:44 -0800
Subject: [PATCH] Move wasm single-thread logic out of test-backend-ops for cpu
 backend

---
 ggml/include/ggml.h            | 6 ++++++
 ggml/src/ggml-cpu/ggml-cpu.cpp | 7 +++++++
 tests/test-backend-ops.cpp     | 4 ----
 3 files changed, 13 insertions(+), 4 deletions(-)

diff --git a/ggml/include/ggml.h b/ggml/include/ggml.h
index 605fcfcb9c..47bfa64536 100644
--- a/ggml/include/ggml.h
+++ b/ggml/include/ggml.h
@@ -225,7 +225,13 @@
 #   define GGML_MAX_NAME        64
 #endif
 
+// For single-thread WASM builds, only use 1 thread
+#if !defined(__EMSCRIPTEN__) || defined(__EMSCRIPTEN_PTHREADS__)
 #define GGML_DEFAULT_N_THREADS  4
+#else
+#define GGML_DEFAULT_N_THREADS  1
+#endif
+
 #define GGML_DEFAULT_GRAPH_SIZE 2048
 
 #if UINTPTR_MAX == 0xFFFFFFFF
diff --git a/ggml/src/ggml-cpu/ggml-cpu.cpp b/ggml/src/ggml-cpu/ggml-cpu.cpp
index 3191faaa4c..3c5cc14102 100644
--- a/ggml/src/ggml-cpu/ggml-cpu.cpp
+++ b/ggml/src/ggml-cpu/ggml-cpu.cpp
@@ -246,8 +246,11 @@ bool ggml_backend_is_cpu(ggml_backend_t backend) {
 void ggml_backend_cpu_set_n_threads(ggml_backend_t backend_cpu, int n_threads) {
     GGML_ASSERT(ggml_backend_is_cpu(backend_cpu));
 
+// For single-thread WASM builds, do not allow changing the number of threads
+#if !defined(_EMSCRIPTEN_) || defined(__EMSCRIPTEN_PTHREADS__)
     struct ggml_backend_cpu_context * ctx = (struct ggml_backend_cpu_context *)backend_cpu->context;
     ctx->n_threads = n_threads;
+#endif
 }
 
 void ggml_backend_cpu_set_threadpool(ggml_backend_t backend_cpu, ggml_threadpool_t threadpool) {
@@ -622,10 +625,14 @@ static ggml_backend_feature * ggml_backend_cpu_get_features(ggml_backend_reg_t r
 }
 
 static void * ggml_backend_cpu_get_proc_address(ggml_backend_reg_t reg, const char * name) {
+
+// For single-thread WASM builds, do not expose a set_n_threads function
+#if !defined(__EMSCRIPTEN__) || defined(__EMSCRIPTEN_PTHREADS__)
     if (strcmp(name, "ggml_backend_set_n_threads") == 0) {
         ggml_backend_set_n_threads_t fct = ggml_backend_cpu_set_n_threads;
         return (void *)fct;
     }
+#endif
     if (strcmp(name, "ggml_backend_dev_get_extra_bufts") == 0) {
         ggml_backend_dev_get_extra_bufts_t fct = ggml_backend_cpu_device_get_extra_buffers_type;
         return (void *)fct;
diff --git a/tests/test-backend-ops.cpp b/tests/test-backend-ops.cpp
index a1df5d6627..5e95e411e9 100644
--- a/tests/test-backend-ops.cpp
+++ b/tests/test-backend-ops.cpp
@@ -19,7 +19,6 @@
 #include <ggml-alloc.h>
 #include <ggml-backend.h>
 #include <ggml-cpp.h>
-#include <ggml-cpu.h>
 
 #include <algorithm>
 #include <array>
@@ -7905,9 +7904,6 @@ static bool test_backend(ggml_backend_t backend, test_mode mode, const char * op
             return false;
         }
 
-        // TODO: find a better way to set the number of threads for the CPU backend
-        ggml_backend_cpu_set_n_threads(backend_cpu, N_THREADS);
-
         size_t n_ok = 0;
         size_t                   tests_run = 0;
         std::vector<std::string> failed_tests;