provide u64 broadcasting mechanism in base layer wavefront lane context, allows more flexible data sharing mechanism which does not depend on statics; reshape dasm building codepath for clicking into artifact cache

src/base/base_entry_point.c

@@ -117,6 +117,7 @@ main_thread_base_entry_point(int arguments_count, char **arguments)
     U64 num_main_threads_clamped = Min(num_async_threads, num_main_threads);
     num_async_threads -= num_main_threads_clamped;
     num_async_threads = Max(1, num_async_threads);
+    U64 lane_broadcast_val = 0;
     Barrier barrier = barrier_alloc(num_async_threads);
     LaneCtx *lane_ctxs = push_array(scratch.arena, LaneCtx, num_async_threads);
     async_threads_count = num_async_threads;
@@ -126,6 +127,7 @@ main_thread_base_entry_point(int arguments_count, char **arguments)
       lane_ctxs[idx].lane_idx = idx;
       lane_ctxs[idx].lane_count = num_async_threads;
       lane_ctxs[idx].barrier = barrier;
+      lane_ctxs[idx].broadcast_memory = &lane_broadcast_val;
       async_threads[idx] = thread_launch(async_thread_entry_point, &lane_ctxs[idx]);
     }
   }

src/base/base_profile.h

@@ -20,6 +20,7 @@
 #if PROFILE_TELEMETRY
 # include "rad_tm.h"
 # if OS_WINDOWS
+#  pragma comment(lib, "ws2_32.lib")
 #  pragma comment(lib, "rad_tm_win64.lib")
 # endif
 #elif PROFILE_SPALL
@@ -44,8 +45,8 @@
 # define ProfLockTake(...)         tmAcquiredLock(0, 0, __VA_ARGS__)
 # define ProfLockDrop(...)         tmReleasedLock(0, __VA_ARGS__)
 # define ProfColor(color)          tmZoneColor((((color) & 0xff000000) >> 24) / 255.f, (((color) & 0x00ff0000) >> 16) / 255.f, (((color) & 0x0000ff00) >> 8) / 255.f)
-# define ProfBeginV(...)                                                           \
-if (TM_API_PTR) {                                                                \
+# define ProfBeginV(...)                                                         \
+if (TM_API_PTR) {                                                               \
 static tm_uint64 file_id = 0; TM_API_PTR->_tmStaticString(&file_id, __FILE__); \
 Temp scratch = scratch_begin(0,0);                                             \
 String8 string = push_str8f(scratch.arena, __VA_ARGS__);                       \
@@ -54,8 +55,8 @@ hash = TM_API_PTR->_tmSendDynamicString(hash, (char*)string.str);              \
 TM_API_PTR->_tmEnterZoneFast_Core(0, 0, file_id, __LINE__, hash);              \
 scratch_end(scratch);                                                          \
 }
-# define ProfNoteV(...) 														   \
-if (TM_API_PTR) {     														   \
+# define ProfNoteV(...) 														                             \
+if (TM_API_PTR) {     														                              \
 static tm_uint64 file_id = 0; TM_API_PTR->_tmStaticString(&file_id, __FILE__); \
 Temp scratch = scratch_begin(0,0);                                             \
 String8 string = push_str8f(scratch.arena, __VA_ARGS__);                       \

src/base/base_thread_context.c

@@ -4,9 +4,9 @@
 ////////////////////////////////
 //~ rjf: Globals
 
-C_LINKAGE thread_static TCTX* tctx_thread_local;
+C_LINKAGE thread_static TCTX *tctx_thread_local;
 #if !BUILD_SUPPLEMENTARY_UNIT
-C_LINKAGE thread_static TCTX* tctx_thread_local = 0;
+C_LINKAGE thread_static TCTX *tctx_thread_local = 0;
 #endif
 
 ////////////////////////////////
@@ -85,12 +85,21 @@ tctx_set_lane_ctx(LaneCtx lane_ctx)
 }
 
 internal void
-tctx_lane_barrier_wait(void)
+tctx_lane_barrier_wait(void *broadcast_ptr, U64 broadcast_size, U64 broadcast_src_lane_idx)
 {
   ProfBeginFunction();
   ProfColor(0x00000ff);
   TCTX *tctx = tctx_selected();
+  U64 broadcast_size_clamped = ClampTop(broadcast_size, sizeof(tctx->lane_ctx.broadcast_memory[0]));
+  if(broadcast_ptr != 0 && lane_idx() == broadcast_src_lane_idx)
+  {
+    MemoryCopy(tctx->lane_ctx.broadcast_memory, broadcast_ptr, broadcast_size_clamped);
+  }
   os_barrier_wait(tctx->lane_ctx.barrier);
+  if(broadcast_ptr != 0 && lane_idx() != broadcast_src_lane_idx)
+  {
+    MemoryCopy(broadcast_ptr, tctx->lane_ctx.broadcast_memory, broadcast_size_clamped);
+  }
   ProfEnd();
 }
 

src/base/base_thread_context.h

@@ -5,7 +5,7 @@
 #define BASE_THREAD_CONTEXT_H
 
 ////////////////////////////////
-//~ rjf: Lane Group Context
+//~ rjf: Lane Context
 
 typedef struct LaneCtx LaneCtx;
 struct LaneCtx
@@ -13,6 +13,7 @@ struct LaneCtx
   U64 lane_idx;
   U64 lane_count;
   Barrier barrier;
+  U64 *broadcast_memory;
 };
 
 ////////////////////////////////
@@ -87,12 +88,13 @@ internal Arena *tctx_get_scratch(Arena **conflicts, U64 count);
 
 //- rjf: lane metadata
 internal LaneCtx tctx_set_lane_ctx(LaneCtx lane_ctx);
-internal void tctx_lane_barrier_wait(void);
+internal void tctx_lane_barrier_wait(void *broadcast_ptr, U64 broadcast_size, U64 broadcast_src_lane_idx);
 #define lane_idx() (tctx_selected()->lane_ctx.lane_idx)
 #define lane_count() (tctx_selected()->lane_ctx.lane_count)
 #define lane_from_task_idx(idx) ((idx)%lane_count())
 #define lane_ctx(ctx) tctx_set_lane_ctx((ctx))
-#define lane_sync() tctx_lane_barrier_wait()
+#define lane_sync() tctx_lane_barrier_wait(0, 0, 0)
+#define lane_sync_u64(ptr, src_lane_idx) tctx_lane_barrier_wait((ptr), sizeof(U64), (src_lane_idx))
 #define lane_range(count) m_range_from_n_idx_m_count(lane_idx(), lane_count(), (count))
 
 //- rjf: thread names