project: sync with upstream

Author: jeffzhou2000

prebuilts/ggml-dsp/README.md

@@ -23,6 +23,6 @@ ggml-hexagon backend consists of two parts:
 
 ### ChangeLog
 
-- 20250531: ggml-hexagon.cpp v1.08 + ggml-dsp v0.96(libggmlop-skel.so, would be removed in the future)
-- 20250607: ggml-hexagon.cpp v1.10 + ggml-dsp v0.96(libggmldsp-skel.so)
-- 20250609: ggml-hexagon.cpp v1.10 + ggml-dsp v0.97(libggmldsp-skel.so)
+- 20250531: ggml-hexagon.cpp v1.08 + ggml-dsp v0.96
+- 20250609: ggml-hexagon.cpp v1.10 + ggml-dsp v0.97
+- 20250625: ggml-hexagon.cpp v1.13 + ggml-dsp v0.98

scripts/build-run-android.sh

@@ -654,6 +654,8 @@ function show_usage()
     echo "  $0 run_benchmark  MUL_MAT       3(cdsp)   mulmat_algotype(0,1,2,3,4,5,6,32,33)  (verify performance of mulmat on cDSP)"
     #verify accuracy    of mulmat on cDSP
     echo "  $0 run_testop     MUL_MAT                 mulmat_algotype(0,1,2,3,4,5,6,32,33)  (verify accuracy    of mulmat on cDSP)"
+    #verify accuracy    of add    on cDSP
+    echo "  $0 run_testop     ADD                                                           (verify accuracy    of add    on cDSP)"
 
     echo -e "\n\n\n"
 }

tests/test-backend-ops.cpp

@@ -4601,6 +4601,13 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
 #endif
 
 #ifdef GGML_USE_HEXAGON
+    //verify computation result of add on cDSP
+    //make whipser.cpp happy
+    test_cases.emplace_back(new test_bin_bcast(ggml_add, GGML_TYPE_F32, {384, 1, 1, 1}, {1,   1, 1, 1}));
+    test_cases.emplace_back(new test_bin_bcast(ggml_add, GGML_TYPE_F32, {1536, 1, 1, 1}, {1,   1, 1, 1}));
+    test_cases.emplace_back(new test_bin_bcast(ggml_add, GGML_TYPE_F32, {2048, 1, 1, 1}, {1,   1, 1, 1}));
+    //verify computation result of fp32 4096x4096 add on cDSP
+    test_cases.emplace_back(new test_bin_bcast(ggml_add, GGML_TYPE_F32, {4096, 4096, 1, 1}, {1,   1, 1, 1}));
     //verify computation result of mulmat on cDSP
     test_cases.emplace_back(new test_mul_mat(GGML_TYPE_F32,    GGML_TYPE_F32, 32, 14, 64, { 1,  1}, {1, 1}));
     test_cases.emplace_back(new test_mul_mat(GGML_TYPE_F32,    GGML_TYPE_F32, 256, 128, 256, { 1,  1}, {1, 1}));