Add RTX 3060 Laptop hw_def; fix CLK_FREQUENCY macro collision

src/cuda/GPU_Microbenchmark/hw_def/ampere_A100_hw_def.h

@@ -12,7 +12,10 @@
 
 #define L1_SIZE (192 * 1024) // Max L1 size in bytes
 
-#define CLK_FREQUENCY 1410 // frequency in MHz
+// NOTE: Frequency commented out since gpuConfig.h defines a `config` struct
+// with a member also named CLK_FREQUENCY, so `#define CLK_FREQUENCY` collides
+
+// #define CLK_FREQUENCY 1410 // frequency in MHz
 
 #define ISSUE_MODEL issue_model::single // single issue core or dual issue
 #define CORE_MODEL core_model::subcore  // subcore model or shared model

src/cuda/GPU_Microbenchmark/hw_def/ampere_RTX3060_MOBILE_hw_def.h

@@ -0,0 +1,36 @@
+// These are the configration parameters that can be found publicly
+// Sources:
+// https://www.nvidia.com/content/dam/en-zz/Solutions/geforce/ampere/pdf/NVIDIA-ampere-GA102-GPU-Architecture-Whitepaper-V1.pdf
+// https://en.wikipedia.org/wiki/GeForce_30_series
+// https://en.wikipedia.org/wiki/CUDA
+
+#ifndef AMPERE_RTX3060_DEF_H
+#define AMPERE_RTX3060_DEF_H
+
+#include "./common/common.h"
+#include "./common/deviceQuery.h"
+
+#define L1_SIZE (128 * 1024) // Max L1 size in bytes
+
+// NOTE: Frequency commented out since gpuConfig.h defines a `config` struct
+// with a member also named CLK_FREQUENCY, so `#define CLK_FREQUENCY` collides
+
+// #define CLK_FREQUENCY 1132 // frequency in MHz
+
+#define ISSUE_MODEL issue_model::single // single issue core or dual issue
+#define CORE_MODEL core_model::subcore  // subcore model or shared model
+#define DRAM_MODEL dram_model::GDDR6    // memory type
+#define WARP_SCHEDS_PER_SM 4            // number of warp schedulers per SM
+
+// number of SASS HMMA per 16x16 PTX WMMA for FP16 - FP32 accumlate operation
+// see slide 22 at
+// https://developer.download.nvidia.com/video/gputechconf/gtc/2020/presentations/s21730-inside-the-nvidia-ampere-architecture.pdf
+#define SASS_hmma_per_PTX_wmma 2
+
+// These vars are almost constant between HW generation
+// see slide 24 from Nvidia at
+// https://developer.download.nvidia.com/video/gputechconf/gtc/2020/presentations/s21730-inside-the-nvidia-ampere-architecture.pdf
+#define L2_BANKS_PER_MEM_CHANNEL 2
+#define L2_BANK_WIDTH_in_BYTE 32
+
+#endif

src/cuda/GPU_Microbenchmark/hw_def/ampere_RTX3070_hw_def.h

@@ -12,7 +12,10 @@
 
 #define L1_SIZE (128 * 1024) // Max L1 size in bytes
 
-#define CLK_FREQUENCY 1132 // frequency in MHz
+// NOTE: Frequency commented out since gpuConfig.h defines a `config` struct
+// with a member also named CLK_FREQUENCY, so `#define CLK_FREQUENCY` collides
+
+// #define CLK_FREQUENCY 1132 // frequency in MHz
 
 #define ISSUE_MODEL issue_model::single // single issue core or dual issue
 #define CORE_MODEL core_model::subcore  // subcore model or shared model

src/cuda/GPU_Microbenchmark/hw_def/blackwell_B200_hw_def.h

@@ -8,6 +8,9 @@
 
 #define L1_SIZE (256 * 1024) // Max L1 size in bytes
 
+// NOTE: Frequency commented out since gpuConfig.h defines a `config` struct
+// with a member also named CLK_FREQUENCY, so `#define CLK_FREQUENCY` collides
+
 // #define CLK_FREQUENCY 1665 // frequency in MHz
 
 #define ISSUE_MODEL issue_model::single // single issue core or dual issue

src/cuda/GPU_Microbenchmark/hw_def/hopper_H100_hw_def.h

@@ -11,7 +11,10 @@
 
 #define L1_SIZE (256 * 1024) // Max L1 size in bytes
 
-#define CLK_FREQUENCY 1455 // frequency in MHz
+// NOTE: Frequency commented out since gpuConfig.h defines a `config` struct
+// with a member also named CLK_FREQUENCY, so `#define CLK_FREQUENCY` collides
+
+// #define CLK_FREQUENCY 1455 // frequency in MHz
 
 #define ISSUE_MODEL issue_model::single
 #define CORE_MODEL core_model::subcore

src/cuda/GPU_Microbenchmark/hw_def/hw_def.h

@@ -11,9 +11,12 @@
 
 //#include "ampere_RTX3070_hw_def.h"
 
-// #include "volta_TITANV_hw_def.h"
+#include "ampere_RTX3060_MOBILE_hw_def.h"
 
-// #include "ampere_A100_hw_def.h"
-#include "blackwell_B200_hw_def.h"
+//#include "volta_TITANV_hw_def.h"
+
+//#include "ampere_A100_hw_def.h"
+
+//#include "blackwell_B200_hw_def.h"
 
 #endif

src/cuda/GPU_Microbenchmark/hw_def/kepler_TITAN_hw_def.h

@@ -7,7 +7,10 @@
 
 #define L1_SIZE (64 * 1024) // Max L1 size in bytes, when enabled
 
-#define CLK_FREQUENCY 837 // frequency in MHz
+// NOTE: Frequency commented out since gpuConfig.h defines a `config` struct
+// with a member also named CLK_FREQUENCY, so `#define CLK_FREQUENCY` collides
+
+// #define CLK_FREQUENCY 837 // frequency in MHz
 
 #define ISSUE_MODEL issue_model::dual
 #define CORE_MODEL core_model::shared

src/cuda/GPU_Microbenchmark/hw_def/pascal_TITANX_hw_def.h

@@ -7,7 +7,10 @@
 
 #define L1_SIZE (24 * 1024) // Max L1 size in bytes, when enabled
 
-#define CLK_FREQUENCY 1417 // frequency in MHz
+// NOTE: Frequency commented out since gpuConfig.h defines a `config` struct
+// with a member also named CLK_FREQUENCY, so `#define CLK_FREQUENCY` collides
+
+//#define CLK_FREQUENCY 1417 // frequency in MHz
 
 #define ISSUE_MODEL issue_model::dual
 #define CORE_MODEL core_model::subcore

src/cuda/GPU_Microbenchmark/hw_def/turing_RTX2060_hw_def.h

@@ -8,7 +8,10 @@
 
 #define L1_SIZE (64 * 1024) // Max L1 size in bytes
 
-#define CLK_FREQUENCY 1365 // frequency in MHz
+// NOTE: Frequency commented out since gpuConfig.h defines a `config` struct
+// with a member also named CLK_FREQUENCY, so `#define CLK_FREQUENCY` collides
+
+// #define CLK_FREQUENCY 1365 // frequency in MHz
 
 #define ISSUE_MODEL issue_model::single   // single issue core or dual issue
 #define CORE_MODEL core_model::subcore    // subcore model or shared model

src/cuda/GPU_Microbenchmark/hw_def/volta_QV100_hw_def.h

@@ -11,7 +11,10 @@
 
 #define L1_SIZE (128 * 1024) // Max L1 size in bytes
 
-#define CLK_FREQUENCY 1132 // frequency in MHz
+// NOTE: Frequency commented out since gpuConfig.h defines a `config` struct
+// with a member also named CLK_FREQUENCY, so `#define CLK_FREQUENCY` collides
+
+// #define CLK_FREQUENCY 1132 // frequency in MHz
 
 #define ISSUE_MODEL issue_model::single
 #define CORE_MODEL core_model::subcore

src/cuda/GPU_Microbenchmark/hw_def/volta_TITANV_hw_def.h

@@ -11,7 +11,10 @@
 
 #define L1_SIZE (128 * 1024) // Max L1 size in bytes
 
-#define CLK_FREQUENCY 1200 // frequency in MHz
+// NOTE: Frequency commented out since gpuConfig.h defines a `config` struct
+// with a member also named CLK_FREQUENCY, so `#define CLK_FREQUENCY` collides
+
+// #define CLK_FREQUENCY 1200 // frequency in MHz
 
 #define ISSUE_MODEL issue_model::single
 #define CORE_MODEL core_model::subcore

src/cuda/GPU_Microbenchmark/hw_def/volta_V100_hw_def.h

@@ -11,7 +11,10 @@
 
 #define L1_SIZE (128 * 1024) // Max L1 size in bytes
 
-#define CLK_FREQUENCY 1455 // frequency in MHz
+// NOTE: Frequency commented out since gpuConfig.h defines a `config` struct
+// with a member also named CLK_FREQUENCY, so `#define CLK_FREQUENCY` collides
+
+// #define CLK_FREQUENCY 1455 // frequency in MHz
 
 #define ISSUE_MODEL issue_model::single
 #define CORE_MODEL core_model::subcore