Fix dispatchKernel arguments in the examples

examples/float16/run.cpp

@@ -46,12 +46,9 @@ int main(int argc, char **argv) {
   }
   Tensor input = createTensor(ctx, Shape{N}, kf16, inputArr.data());
   Tensor output = createTensor(ctx, Shape{N}, kf16);
-  std::promise<void> promise;
-  std::future<void> future = promise.get_future();
   Kernel op = createKernel(ctx, {kGelu, 256, kf16}, Bindings{input, output},
                            {cdiv(N, 256), 1, 1});
-  dispatchKernel(ctx, op, promise);
-  wait(ctx, future);
+  dispatchKernel(ctx, op);
   toCPU(ctx, output, outputArr.data(), sizeof(outputArr));
 
   for (int i = 0; i < 12; ++i) {

examples/gpu_puzzles/run.cpp

@@ -23,11 +23,8 @@ template <size_t N> std::array<float, N> makeData() {
 
 template <size_t N, size_t R = N, size_t C = 1> void showResult(Context &ctx, Kernel &op, Tensor &output) {
 
-  std::promise<void> promise;
-  std::future<void> future = promise.get_future();
-  dispatchKernel(ctx, op, promise);
+  dispatchKernel(ctx, op);
   std::array<float, R * C> outputArr;
-  wait(ctx, future);
   toCPU(ctx, output, outputArr.data(), sizeof(outputArr));
   printf("%s", show<float, R, C>(outputArr, "output").c_str());
 }

examples/matmul/run.cpp

@@ -838,12 +838,9 @@ void runTest(int version, size_t M, size_t K, size_t N,
 
   // Initialize Kernel and bind GPU buffers
   // pre-allocate for async dispatch
-  std::array<std::promise<void>, nIter> promises;
-  std::array<std::future<void>, nIter> futures;
   std::array<Kernel, nIter> kernels;
   std::array<Tensor, nIter> outputs;
   for (int i = 0; i < nIter; i++) {
-    futures[i] = promises[i].get_future();
     outputs[i] = createTensor(ctx, Shape{M, N}, numtype);
     kernels[i] = selectMatmul(ctx, version, {input, weights, outputs[i]}, M, K, N, numtype);
   }
@@ -854,10 +851,7 @@ void runTest(int version, size_t M, size_t K, size_t N,
   // Dispatch kernel nIter times
   auto start = std::chrono::high_resolution_clock::now();
   for (int i = 0; i < nIter; i++) {
-    dispatchKernel(ctx, kernels[i], promises[i]);
-  }
-  for (int i = 0; i < nIter; i++) {
-    wait(ctx, futures[i]);
+    dispatchKernel(ctx, kernels[i]);
   }
   auto end = std::chrono::high_resolution_clock::now();
 

examples/physics/run.cpp

@@ -84,10 +84,7 @@ int main() {
   printf("\033[2J\033[H");
   while (true) {
     auto start = std::chrono::high_resolution_clock::now();
-    std::promise<void> promise;
-    std::future<void> future = promise.get_future();
-    dispatchKernel(ctx, update, promise);
-    wait(ctx, future);
+    dispatchKernel(ctx, update);
     toCPU(ctx, pos, posArr.data(), sizeof(posArr));
     auto end = std::chrono::high_resolution_clock::now();
     std::chrono::duration<double> elapsed = end - start;

examples/shadertui/run.cpp

@@ -126,10 +126,7 @@ int main() {
     params.time = getCurrentTimeInMilliseconds(start);
     toGPU(ctx, params, renderKernel);
     auto frameStart = std::chrono::high_resolution_clock::now();
-    std::promise<void> promise;
-    std::future<void> future = promise.get_future();
-    dispatchKernel(ctx, renderKernel, promise);
-    wait(ctx, future);
+    dispatchKernel(ctx, renderKernel);
     resetCommandBuffer(ctx.device, renderKernel);
     toCPU(ctx, screen, screenArr);
     rasterize<kRows, kCols>(screenArr, raster);

examples/transpose/run.cpp

@@ -162,20 +162,11 @@ void runTest(int version, size_t M, size_t N,
   LOG(kDefLog, kInfo, "Dispatching Kernel version %d, %d iterations ...",
       version, nIter);
 
-  // pre-allocate promises and futures for async dispatch
-  // TODO(avh): implement a pooling mechanism for promises/futures in gpu.h
-  std::array<std::promise<void>, nIter> promises;
-  std::array<std::future<void>, nIter> futures;
-  for (int i = 0; i < nIter; i++) {
-    futures[i] = promises[i].get_future();
-  }
-
   // Dispatch kernel nIter times
   auto start = std::chrono::high_resolution_clock::now();
   for (int i = 0; i < nIter; i++) {
     if (!isCPU) {
-      dispatchKernel(ctx, kernel, promises[i]);
-      wait(ctx, futures[i]);
+      dispatchKernel(ctx, kernel);
       resetCommandBuffer(ctx.device, kernel);
     } else {
       transpose(inputPtr.get(), outputPtr.get(), M, N);