Move su_load_gmdls into the responsibility of the intro to call, anticipating multicore rendering so it is called once before all the cores are spun up.

src/sointu.asm

@@ -339,9 +339,6 @@ SECT_TEXT(surender)
 
 EXPORT MANGLE_FUNC(su_render,PTRSIZE)   ; Stack: ptr
     render_prologue
-%ifdef INCLUDE_GMDLS
-    call    su_gmdls_load
-%endif
     xor     eax, eax
 %ifdef INCLUDE_MULTIVOICE_TRACKS
     push    VOICETRACK_BITMASK

src/win32/gmdls_win32.asm

@@ -7,21 +7,21 @@ extern _ReadFile@20 ; requires windows
 
 SECT_TEXT(sugmdls)
 
-su_gmdls_load:
+EXPORT MANGLE_FUNC(su_load_gmdls,0)
-    mov     edi, MANGLE_DATA(su_sample_table)
+    mov     edx, MANGLE_DATA(su_sample_table)
-    mov     esi, su_gmdls_path1
+    mov     ecx, su_gmdls_path1
     su_gmdls_pathloop:
         push    0                   ; OF_READ
-        push    edi                 ; &ofstruct, blatantly reuse the sample table
+        push    edx                 ; &ofstruct, blatantly reuse the sample table
-        push    esi                 ; path
+        push    ecx                 ; path
         call    _OpenFile@12        ; eax = OpenFile(path,&ofstruct,OF_READ)
-        add     esi, su_gmdls_path2 - su_gmdls_path1 ; if we ever get to third, then crash
+        add     ecx, su_gmdls_path2 - su_gmdls_path1 ; if we ever get to third, then crash
         cmp     eax, -1             ; eax == INVALID?
         je      su_gmdls_pathloop
     push    0                       ; NULL
-    push    edi                     ; &bytes_read, reusing sample table again; it does not matter that the first four bytes are trashed
+    push    edx                     ; &bytes_read, reusing sample table again; it does not matter that the first four bytes are trashed
     push    SAMPLE_TABLE_SIZE       ; number of bytes to read
-    push    edi                     ; here we actually pass the sample table to readfile
+    push    edx                     ; here we actually pass the sample table to readfile
     push    eax                     ; handle to file
     call    _ReadFile@20            ; Readfile(handle,&su_sample_table,SAMPLE_TABLE_SIZE,&bytes_read,NULL)
     ret

src/win64/gmdls_win64.asm

@@ -6,24 +6,26 @@ extern OpenFile ; requires windows
 extern ReadFile ; requires windows
 
 SECT_TEXT(sugmdls)
+
+EXPORT MANGLE_FUNC(su_load_gmdls,0)
 ;        Win64 ABI: RCX, RDX, R8, and R9
-su_gmdls_load:
     sub     rsp, 40         ; Win64 ABI requires "shadow space" + space for one parameter.
-    mov     rdi, PTRWORD MANGLE_DATA(su_sample_table)
+    mov     rdx, PTRWORD MANGLE_DATA(su_sample_table)
-    mov     rsi, PTRWORD su_gmdls_path1
+    mov     rcx, PTRWORD su_gmdls_path1
     su_gmdls_pathloop:
         xor     r8,r8 ; OF_READ
-        mov     rdx, rdi             ; &ofstruct, blatantly reuse the sample table
+        push    rdx                 ; &ofstruct, blatantly reuse the sample table
-        mov     rcx, rsi        ; path
+        push    rcx
         call    OpenFile            ; eax = OpenFile(path,&ofstruct,OF_READ)
-        add     rsi, su_gmdls_path2 - su_gmdls_path1 ; if we ever get to third, then crash
+        pop     rcx
-        movsxd  rcx,eax
+        add     rcx, su_gmdls_path2 - su_gmdls_path1 ; if we ever get to third, then crash
-        cmp     rcx, -1             ; ecx == INVALID?
+        pop     rdx
+        cmp     eax, -1             ; ecx == INVALID?
         je      su_gmdls_pathloop
+    movsxd  rcx, eax
     mov     qword [rsp+32],0
-    mov     r9, rdi
+    mov     r9, rdx
     mov     r8d, SAMPLE_TABLE_SIZE   ; number of bytes to read
-    mov     rdx, rdi
     call    ReadFile                ; Readfile(handle,&su_sample_table,SAMPLE_TABLE_SIZE,&bytes_read,NULL)
     add     rsp, 40         ; shadow space, as required by Win64 ABI
     ret

tests/CMakeLists.txt

@@ -93,7 +93,9 @@ regression_test(test_oscillat_gate ENVELOPE)
 regression_test(test_oscillat_stereo ENVELOPE)
 if(WIN32) # The samples are currently only GMDLs based, and thus require Windows.
     regression_test(test_oscillat_sample ENVELOPE)
+    target_compile_definitions(test_oscillat_sample PUBLIC INCLUDE_GMDLS)
     regression_test(test_oscillat_sample_stereo ENVELOPE)
+    target_compile_definitions(test_oscillat_sample_stereo PUBLIC INCLUDE_GMDLS)
 endif()
 regression_test(test_oscillat_unison ENVELOPE)
 regression_test(test_oscillat_unison_stereo ENVELOPE)

tests/test_oscillat_sample.asm

@@ -1,5 +1,4 @@
 %define BPM 100
-%define INCLUDE_GMDLS
 
 %include "../src/sointu.inc"
 

tests/test_oscillat_sample_stereo.asm

@@ -1,5 +1,4 @@
 %define BPM 100
-%define INCLUDE_GMDLS
 
 %include "../src/sointu.inc"
 

tests/test_renderer.c

@@ -26,6 +26,10 @@
 #endif
 extern void CALLCONV su_render(void *);
 
+#ifdef INCLUDE_GMDLS
+extern void CALLCONV su_load_gmdls(void);
+#endif
+
 extern int su_max_samples;
 
 int main(int argc, char* argv[]) {
@@ -59,6 +63,10 @@ int main(int argc, char* argv[]) {
         return 1;
     }
 
+    #ifdef INCLUDE_GMDLS
+    su_load_gmdls();
+    #endif
+
     su_render(buf);
 
     snprintf(filename, sizeof filename, "%s%s%s", expected_output_folder, test_name, ".raw");