#!amber
# Copyright 2024 The Amber Authors.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.


# This benchmark tests the latency and throughput of
# shared memory vs  ssbo (main cached) memory.
# Configs to manually modify:
#  - Comment in/out declaration : Shared vs ssbo
#  - local_size_x (workgroup size) : Single SM throughput
#  - number of loop unrolls : latency of single thread
#  - compute/dispatch size (currently 1) : Device throughput


SHADER compute workgroup_shared_vs_ssbo GLSL
#version 430

layout(local_size_x = 1, local_size_y = 1, local_size_z = 1) in;

// Comment in/out these two lines to test shared memory
struct BlockB { uint data[8]; };  shared BlockB ssbo_wr;
//layout(set = 0, binding = 0) buffer BlockB { uint data[];} ssbo_wr;

layout(set = 0, binding = 1) buffer BlockA { 
  uint data[];
} ssbo_fake_volatile;


void main() {
    // This is required when using shared memory
    if( gl_LocalInvocationID.x == 0){
      ssbo_wr.data[0] = 0;
    }
    barrier();
    uint fv = ssbo_fake_volatile.data[0];
    uint iter_val = ssbo_wr.data[fv];
    for(uint i = 0;i<1000;i++){
      // 10x
      iter_val = ssbo_wr.data[iter_val];
      iter_val = ssbo_wr.data[iter_val];
      iter_val = ssbo_wr.data[iter_val];
      iter_val = ssbo_wr.data[iter_val];
      iter_val = ssbo_wr.data[iter_val];
      iter_val = ssbo_wr.data[iter_val];
      iter_val = ssbo_wr.data[iter_val];
      iter_val = ssbo_wr.data[iter_val];
      iter_val = ssbo_wr.data[iter_val];
      iter_val = ssbo_wr.data[iter_val];

      // 10x
      iter_val = ssbo_wr.data[iter_val];
      iter_val = ssbo_wr.data[iter_val];
      iter_val = ssbo_wr.data[iter_val];
      iter_val = ssbo_wr.data[iter_val];
      iter_val = ssbo_wr.data[iter_val];
      iter_val = ssbo_wr.data[iter_val];
      iter_val = ssbo_wr.data[iter_val];
      iter_val = ssbo_wr.data[iter_val];
      iter_val = ssbo_wr.data[iter_val];
      iter_val = ssbo_wr.data[iter_val];
    }
    ssbo_wr.data[gl_LocalInvocationID.x]  = iter_val;
}
END

BUFFER buf_uint DATA_TYPE uint32 SIZE 1024 FILL 0
BUFFER buf_fake_volatile DATA_TYPE uint32 SIZE 1048576 FILL 0

PIPELINE compute pipeline
  ATTACH workgroup_shared_vs_ssbo
  BIND BUFFER buf_uint AS storage DESCRIPTOR_SET 0 BINDING 0
  BIND BUFFER buf_fake_volatile AS storage DESCRIPTOR_SET 0 BINDING 1
END

REPEAT 333
RUN TIMED_EXECUTION pipeline 1 1 1
END