Added prototype for ixAMAX routines

10 files changed, 256 insertions, 2 deletions

diff --git a/include/clblast.h b/include/clblast.h
index cf55a256..57948581 100644
--- a/include/clblast.h
+++ b/include/clblast.h
@@ -188,6 +188,13 @@ StatusCode Asum(const size_t n,
                 const cl_mem x_buffer, const size_t x_offset, const size_t x_inc,
                 cl_command_queue* queue, cl_event* event = nullptr);
 
+// Index of absolute maxium value in a vector: iSAMAX/iDAMAX/iCAMAX/iZAMAX
+template <typename T>
+StatusCode Amax(const size_t n,
+                cl_mem imax_buffer, const size_t imax_offset,
+                const cl_mem x_buffer, const size_t x_offset, const size_t x_inc,
+                cl_command_queue* queue, cl_event* event = nullptr);
+
 // =================================================================================================
 // BLAS level-2 (matrix-vector) routines
 // =================================================================================================
diff --git a/include/clblast_c.h b/include/clblast_c.h
index 190e6a46..92f4afe5 100644
--- a/include/clblast_c.h
+++ b/include/clblast_c.h
@@ -296,6 +296,24 @@ StatusCode PUBLIC_API CLBlastDzasum(const size_t n,
                                    const cl_mem x_buffer, const size_t x_offset, const size_t x_inc,
                                    cl_command_queue* queue, cl_event* event);
 
+// Index of absolute maxium value in a vector: iSAMAX/iDAMAX/iCAMAX/iZAMAX
+StatusCode PUBLIC_API CLBlastiSamax(const size_t n,
+                                   cl_mem imax_buffer, const size_t imax_offset,
+                                   const cl_mem x_buffer, const size_t x_offset, const size_t x_inc,
+                                   cl_command_queue* queue, cl_event* event);
+StatusCode PUBLIC_API CLBlastiDamax(const size_t n,
+                                   cl_mem imax_buffer, const size_t imax_offset,
+                                   const cl_mem x_buffer, const size_t x_offset, const size_t x_inc,
+                                   cl_command_queue* queue, cl_event* event);
+StatusCode PUBLIC_API CLBlastiCamax(const size_t n,
+                                   cl_mem imax_buffer, const size_t imax_offset,
+                                   const cl_mem x_buffer, const size_t x_offset, const size_t x_inc,
+                                   cl_command_queue* queue, cl_event* event);
+StatusCode PUBLIC_API CLBlastiZamax(const size_t n,
+                                   cl_mem imax_buffer, const size_t imax_offset,
+                                   const cl_mem x_buffer, const size_t x_offset, const size_t x_inc,
+                                   cl_command_queue* queue, cl_event* event);
+
 // =================================================================================================
 // BLAS level-2 (matrix-vector) routines
 // =================================================================================================
diff --git a/scripts/generator/generator.py b/scripts/generator/generator.py
index 9b1b2aa1..3a845686 100644
--- a/scripts/generator/generator.py
+++ b/scripts/generator/generator.py
@@ -40,6 +40,10 @@ Z = DataType("Z", "Z", DBL2, [DBL2, DBL2, D2CL, D2CL], DBL2) # double-complex (6
 # Special cases
 Sc = DataType("C", "Sc", FLT2,         [FLT2, FLT2, FLT2, FLT2], FLT2) # As C, but with real output
 Dz = DataType("Z", "Dz", DBL2,         [DBL2, DBL2, DBL2, DBL2], DBL2) # As Z, but with real output
+iS = DataType("S", "iS", FLT,          [FLT,  FLT,  FLT,  FLT],  FLT ) # As S, but with integer output
+iD = DataType("D", "iD", DBL,          [DBL,  DBL,  DBL,  DBL],  DBL ) # As D, but with integer output
+iC = DataType("C", "iC", FLT2,         [FLT2, FLT2, F2CL, F2CL], FLT2) # As C, but with integer output
+iZ = DataType("Z", "iZ", DBL2,         [DBL2, DBL2, D2CL, D2CL], DBL2) # As Z, but with integer output
 Css = DataType("C", "C", FLT,          [FLT,  FLT,  FLT,  FLT], FLT2) # As C, but with constants from S
 Zdd = DataType("Z", "Z", DBL,          [DBL,  DBL,  DBL,  DBL], DBL2) # As Z, but with constants from D
 Ccs = DataType("C", "C", FLT2+","+FLT, [FLT2, FLT,  F2CL, FLT], FLT2) # As C, but with one constant from S
@@ -68,6 +72,7 @@ routines = [
   Routine(True,  "1", "dotc",  T,  [C,Z],     ["n"], [], ["x","y"], ["dot"], [], "n", "Dot product of two complex vectors, one conjugated"),
   Routine(True,  "1", "nrm2",  T, [S,D,Sc,Dz],["n"], [], ["x"], ["nrm2"], [], "2*n", "Euclidian norm of a vector"),
   Routine(True,  "1", "asum",  T, [S,D,Sc,Dz],["n"], [], ["x"], ["asum"], [], "n", "Absolute sum of values in a vector"),
+  Routine(False, "1", "amax",  T, [iS,iD,iC,iZ],["n"], [], ["x"], ["imax"], [], "n", "Index of absolute maxium value in a vector"),
 ],
 [ # Level 2: matrix-vector
   Routine(True,  "2a", "gemv",  T,  [S,D,C,Z], ["m","n"], ["layout","a_transpose"], ["a","x"], ["y"], ["alpha","beta"], "", "General matrix-vector multiplication"),
@@ -369,7 +374,7 @@ for level in [1,2,3]:
 			body += "using double2 = clblast::double2;\n\n"
 			body += "// Main function (not within the clblast namespace)\n"
 			body += "int main(int argc, char *argv[]) {\n"
-			default = PrecisionToFullName(routine.flavours[0].name)
+			default = PrecisionToFullName(routine.flavours[0].precision_name)
 			body += "  switch(clblast::GetPrecision(argc, argv, clblast::Precision::k"+default+")) {\n"
 			for precision in ["H","S","D","C","Z"]:
 				body += "    case clblast::Precision::k"+PrecisionToFullName(precision)+":"
diff --git a/scripts/generator/routine.py b/scripts/generator/routine.py
index 7ddd7a12..9806d960 100644
--- a/scripts/generator/routine.py
+++ b/scripts/generator/routine.py
@@ -72,7 +72,7 @@ class Routine():
 
 	# List of scalar buffers
 	def ScalarBuffersFirst(self):
-		return ["dot","nrm2","asum"]
+		return ["dot","nrm2","asum","imax"]
 	def ScalarBuffersSecond(self):
 		return ["sa","sb","sc","ss","sd1","sd2","sx1","sy1","sparam"]
 
diff --git a/src/clblast.cc b/src/clblast.cc
index 7210ad1d..bee63b53 100644
--- a/src/clblast.cc
+++ b/src/clblast.cc
@@ -428,6 +428,31 @@ template StatusCode PUBLIC_API Asum<double2>(const size_t,
                                              const cl_mem, const size_t, const size_t,
                                              cl_command_queue*, cl_event*);
 
+// Index of absolute maxium value in a vector: iSAMAX/iDAMAX/iCAMAX/iZAMAX
+template <typename T>
+StatusCode Amax(const size_t,
+                cl_mem, const size_t,
+                const cl_mem, const size_t, const size_t,
+                cl_command_queue*, cl_event*) {
+  return StatusCode::kNotImplemented;
+}
+template StatusCode PUBLIC_API Amax<float>(const size_t,
+                                           cl_mem, const size_t,
+                                           const cl_mem, const size_t, const size_t,
+                                           cl_command_queue*, cl_event*);
+template StatusCode PUBLIC_API Amax<double>(const size_t,
+                                            cl_mem, const size_t,
+                                            const cl_mem, const size_t, const size_t,
+                                            cl_command_queue*, cl_event*);
+template StatusCode PUBLIC_API Amax<float2>(const size_t,
+                                            cl_mem, const size_t,
+                                            const cl_mem, const size_t, const size_t,
+                                            cl_command_queue*, cl_event*);
+template StatusCode PUBLIC_API Amax<double2>(const size_t,
+                                             cl_mem, const size_t,
+                                             const cl_mem, const size_t, const size_t,
+                                             cl_command_queue*, cl_event*);
+
 // =================================================================================================
 // BLAS level-2 (matrix-vector) routines
 // =================================================================================================
diff --git a/src/clblast_c.cc b/src/clblast_c.cc
index c36edbca..23c96feb 100644
--- a/src/clblast_c.cc
+++ b/src/clblast_c.cc
@@ -475,6 +475,48 @@ StatusCode CLBlastDzasum(const size_t n,
   return static_cast<StatusCode>(status);
 }
 
+// AMAX
+StatusCode CLBlastiSamax(const size_t n,
+                        cl_mem imax_buffer, const size_t imax_offset,
+                        const cl_mem x_buffer, const size_t x_offset, const size_t x_inc,
+                        cl_command_queue* queue, cl_event* event) {
+  auto status = clblast::Amax<float>(n,
+                                     imax_buffer, imax_offset,
+                                     x_buffer, x_offset, x_inc,
+                                     queue, event);
+  return static_cast<StatusCode>(status);
+}
+StatusCode CLBlastiDamax(const size_t n,
+                        cl_mem imax_buffer, const size_t imax_offset,
+                        const cl_mem x_buffer, const size_t x_offset, const size_t x_inc,
+                        cl_command_queue* queue, cl_event* event) {
+  auto status = clblast::Amax<double>(n,
+                                      imax_buffer, imax_offset,
+                                      x_buffer, x_offset, x_inc,
+                                      queue, event);
+  return static_cast<StatusCode>(status);
+}
+StatusCode CLBlastiCamax(const size_t n,
+                        cl_mem imax_buffer, const size_t imax_offset,
+                        const cl_mem x_buffer, const size_t x_offset, const size_t x_inc,
+                        cl_command_queue* queue, cl_event* event) {
+  auto status = clblast::Amax<float2>(n,
+                                      imax_buffer, imax_offset,
+                                      x_buffer, x_offset, x_inc,
+                                      queue, event);
+  return static_cast<StatusCode>(status);
+}
+StatusCode CLBlastiZamax(const size_t n,
+                        cl_mem imax_buffer, const size_t imax_offset,
+                        const cl_mem x_buffer, const size_t x_offset, const size_t x_inc,
+                        cl_command_queue* queue, cl_event* event) {
+  auto status = clblast::Amax<double2>(n,
+                                       imax_buffer, imax_offset,
+                                       x_buffer, x_offset, x_inc,
+                                       queue, event);
+  return static_cast<StatusCode>(status);
+}
+
 // =================================================================================================
 // BLAS level-2 (matrix-vector) routines
 // =================================================================================================
diff --git a/test/correctness/routines/level1/xamax.cc b/test/correctness/routines/level1/xamax.cc
new file mode 100644
index 00000000..ade09e7a
--- /dev/null
+++ b/test/correctness/routines/level1/xamax.cc
@@ -0,0 +1,28 @@
+
+// =================================================================================================
+// This file is part of the CLBlast project. The project is licensed under Apache Version 2.0. This
+// project loosely follows the Google C++ styleguide and uses a tab-size of two spaces and a max-
+// width of 100 characters per line.
+//
+// Author(s):
+//   Cedric Nugteren <www.cedricnugteren.nl>
+//
+// =================================================================================================
+
+#include "correctness/testblas.h"
+#include "routines/level1/xamax.h"
+
+// Shortcuts to the clblast namespace
+using float2 = clblast::float2;
+using double2 = clblast::double2;
+
+// Main function (not within the clblast namespace)
+int main(int argc, char *argv[]) {
+  clblast::RunTests<clblast::TestXamax<float>, float, float>(argc, argv, false, "iSAMAX");
+  clblast::RunTests<clblast::TestXamax<double>, double, double>(argc, argv, true, "iDAMAX");
+  clblast::RunTests<clblast::TestXamax<float2>, float2, float2>(argc, argv, true, "iCAMAX");
+  clblast::RunTests<clblast::TestXamax<double2>, double2, double2>(argc, argv, true, "iZAMAX");
+  return 0;
+}
+
+// =================================================================================================
diff --git a/test/performance/routines/level1/xamax.cc b/test/performance/routines/level1/xamax.cc
new file mode 100644
index 00000000..85caa483
--- /dev/null
+++ b/test/performance/routines/level1/xamax.cc
@@ -0,0 +1,35 @@
+
+// =================================================================================================
+// This file is part of the CLBlast project. The project is licensed under Apache Version 2.0. This
+// project loosely follows the Google C++ styleguide and uses a tab-size of two spaces and a max-
+// width of 100 characters per line.
+//
+// Author(s):
+//   Cedric Nugteren <www.cedricnugteren.nl>
+//
+// =================================================================================================
+
+#include "performance/client.h"
+#include "routines/level1/xamax.h"
+
+// Shortcuts to the clblast namespace
+using float2 = clblast::float2;
+using double2 = clblast::double2;
+
+// Main function (not within the clblast namespace)
+int main(int argc, char *argv[]) {
+  switch(clblast::GetPrecision(argc, argv, clblast::Precision::kSingle)) {
+    case clblast::Precision::kHalf: throw std::runtime_error("Unsupported precision mode");
+    case clblast::Precision::kSingle:
+      clblast::RunClient<clblast::TestXamax<float>, float, float>(argc, argv); break;
+    case clblast::Precision::kDouble:
+      clblast::RunClient<clblast::TestXamax<double>, double, double>(argc, argv); break;
+    case clblast::Precision::kComplexSingle:
+      clblast::RunClient<clblast::TestXamax<float2>, float2, float2>(argc, argv); break;
+    case clblast::Precision::kComplexDouble:
+      clblast::RunClient<clblast::TestXamax<double2>, double2, double2>(argc, argv); break;
+  }
+  return 0;
+}
+
+// =================================================================================================
diff --git a/test/wrapper_cblas.h b/test/wrapper_cblas.h
index af0eec9b..994b48b1 100644
--- a/test/wrapper_cblas.h
+++ b/test/wrapper_cblas.h
@@ -371,6 +371,32 @@ void cblasXasum(const size_t n,
                                          reinterpret_cast<const double*>(&x_buffer[x_offset]), static_cast<int>(x_inc));
 }
 
+// Forwards the Netlib BLAS calls for iSAMAX/iDAMAX/iCAMAX/iZAMAX
+void cblasXamax(const size_t n,
+                std::vector<float>& imax_buffer, const size_t imax_offset,
+                const std::vector<float>& x_buffer, const size_t x_offset, const size_t x_inc) {
+  imax_buffer[imax_offset] = cblas_isamax(n,
+                                         &x_buffer[x_offset], static_cast<int>(x_inc));
+}
+void cblasXamax(const size_t n,
+                std::vector<double>& imax_buffer, const size_t imax_offset,
+                const std::vector<double>& x_buffer, const size_t x_offset, const size_t x_inc) {
+  imax_buffer[imax_offset] = cblas_idamax(n,
+                                         &x_buffer[x_offset], static_cast<int>(x_inc));
+}
+void cblasXamax(const size_t n,
+                std::vector<float2>& imax_buffer, const size_t imax_offset,
+                const std::vector<float2>& x_buffer, const size_t x_offset, const size_t x_inc) {
+  imax_buffer[imax_offset] = cblas_icamax(n,
+                                         reinterpret_cast<const float*>(&x_buffer[x_offset]), static_cast<int>(x_inc));
+}
+void cblasXamax(const size_t n,
+                std::vector<double2>& imax_buffer, const size_t imax_offset,
+                const std::vector<double2>& x_buffer, const size_t x_offset, const size_t x_inc) {
+  imax_buffer[imax_offset] = cblas_izamax(n,
+                                         reinterpret_cast<const double*>(&x_buffer[x_offset]), static_cast<int>(x_inc));
+}
+
 // =================================================================================================
 // BLAS level-2 (matrix-vector) routines
 // =================================================================================================
diff --git a/test/wrapper_clblas.h b/test/wrapper_clblas.h
index 09b3310b..955dc3ad 100644
--- a/test/wrapper_clblas.h
+++ b/test/wrapper_clblas.h
@@ -626,6 +626,74 @@ clblasStatus clblasXasum<double2>(const size_t n,
                      num_queues, queues, num_wait_events, wait_events, events);
 }
 
+// Forwards the clBLAS calls for iSAMAX/iDAMAX/iCAMAX/iZAMAX
+template <typename T>
+clblasStatus clblasXamax(const size_t n,
+                         cl_mem imax_buffer, const size_t imax_offset,
+                         const cl_mem x_buffer, const size_t x_offset, const size_t x_inc,
+                         cl_uint num_queues, cl_command_queue *queues,
+                         cl_uint num_wait_events, const cl_event *wait_events, cl_event *events);
+template <>
+clblasStatus clblasXamax<float>(const size_t n,
+                                cl_mem imax_buffer, const size_t imax_offset,
+                                const cl_mem x_buffer, const size_t x_offset, const size_t x_inc,
+                                cl_uint num_queues, cl_command_queue *queues,
+                                cl_uint num_wait_events, const cl_event *wait_events, cl_event *events) {
+  auto queue = Queue(queues[0]);
+  auto context = queue.GetContext();
+  auto scratch_buffer = Buffer<float>(context, n);
+  return clblasiSamax(n,
+                     imax_buffer, imax_offset,
+                     x_buffer, x_offset, static_cast<int>(x_inc),
+                     scratch_buffer(),
+                     num_queues, queues, num_wait_events, wait_events, events);
+}
+template <>
+clblasStatus clblasXamax<double>(const size_t n,
+                                 cl_mem imax_buffer, const size_t imax_offset,
+                                 const cl_mem x_buffer, const size_t x_offset, const size_t x_inc,
+                                 cl_uint num_queues, cl_command_queue *queues,
+                                 cl_uint num_wait_events, const cl_event *wait_events, cl_event *events) {
+  auto queue = Queue(queues[0]);
+  auto context = queue.GetContext();
+  auto scratch_buffer = Buffer<double>(context, n);
+  return clblasiDamax(n,
+                     imax_buffer, imax_offset,
+                     x_buffer, x_offset, static_cast<int>(x_inc),
+                     scratch_buffer(),
+                     num_queues, queues, num_wait_events, wait_events, events);
+}
+template <>
+clblasStatus clblasXamax<float2>(const size_t n,
+                                 cl_mem imax_buffer, const size_t imax_offset,
+                                 const cl_mem x_buffer, const size_t x_offset, const size_t x_inc,
+                                 cl_uint num_queues, cl_command_queue *queues,
+                                 cl_uint num_wait_events, const cl_event *wait_events, cl_event *events) {
+  auto queue = Queue(queues[0]);
+  auto context = queue.GetContext();
+  auto scratch_buffer = Buffer<float2>(context, n);
+  return clblasiCamax(n,
+                     imax_buffer, imax_offset,
+                     x_buffer, x_offset, static_cast<int>(x_inc),
+                     scratch_buffer(),
+                     num_queues, queues, num_wait_events, wait_events, events);
+}
+template <>
+clblasStatus clblasXamax<double2>(const size_t n,
+                                  cl_mem imax_buffer, const size_t imax_offset,
+                                  const cl_mem x_buffer, const size_t x_offset, const size_t x_inc,
+                                  cl_uint num_queues, cl_command_queue *queues,
+                                  cl_uint num_wait_events, const cl_event *wait_events, cl_event *events) {
+  auto queue = Queue(queues[0]);
+  auto context = queue.GetContext();
+  auto scratch_buffer = Buffer<double2>(context, n);
+  return clblasiZamax(n,
+                     imax_buffer, imax_offset,
+                     x_buffer, x_offset, static_cast<int>(x_inc),
+                     scratch_buffer(),
+                     num_queues, queues, num_wait_events, wait_events, events);
+}
+
 // =================================================================================================
 // BLAS level-2 (matrix-vector) routines
 // =================================================================================================