type definitions and rotm

.github/workflows/tests.yml

@@ -1,9 +1,12 @@
 name: tests
 
-on: [push]
+on: 
+  push:
+      branches:
+          - main
 
 jobs:
-  no_ext_tests:
+  tests:
     runs-on: ubuntu-22.04
 
     strategy:
@@ -21,6 +24,12 @@ jobs:
           php_version: ${{ matrix.php }}
           php_extensions: ffi
 
+      - name: PHP Static Analysys
+        uses: php-actions/phpstan@v3
+        with:
+          php_version: ${{ matrix.php }}
+          path: src/
+
       - name: PHPUnit Tests
         uses: php-actions/phpunit@v3
         with:

README.md

@@ -51,14 +51,25 @@ Ubuntu standard OpenCL drivers include:
 - pocl-opencl-icd
 
 
-Next, download and set up clbast.
+Next, Setup clblast. 
 
-Download and Extract Archive file and Pack to deb
+Install clbast on Ubuntu 22.04 or Debian 12 or later.
+```shell
+$ sudo apt install libclblast1
+```
+
+If you use Ubuntu 20.04 or Debian 11,
+download and Extract Archive file and Pack to deb
 ```shell
 $ cd /some/app/directory
 $ composer require rindow/rindow-clblast-ffi
 $ cp vendor/rindow/rindow-clblast-ffi/clblast-packdeb.sh .
 $ sh ./clblast-packdeb.sh
-$ sudo apt install ./clblast_X.X.X-X+ubuntuXX.XX_amd64.deb
+$ sudo apt install ./clblast_X.X.X_amd64.deb
 ```
 
+And then, Please install rindow-clblast-ffi if you have not already done so.
+```shell
+$ cd /some/app/directory
+$ composer require rindow/rindow-clblast-ffi
+```

clblast-packdeb.sh

@@ -1,15 +1,15 @@
 #!/usr/bin/bash
 
 CLBLASTVERSION=1.6.2
-
-. /etc/os-release
+#CLBLASTVERSION=1.5.2
 
 FILENAME=CLBlast-${CLBLASTVERSION}-linux-x86_64
-PLATFORM=ubuntu${VERSION_ID}
+#FILENAME=CLBlast-${CLBLASTVERSION}-Linux-x64
 TARGET=./pkgwork
 
 
 wget https://github.com/CNugteren/CLBlast/releases/download/${CLBLASTVERSION}/${FILENAME}.zip
+#wget https://github.com/CNugteren/CLBlast/releases/download/${CLBLASTVERSION}/${FILENAME}.tar.xz
 
 unzip ${FILENAME}.zip
 tar xvf ${FILENAME}.tar.gz
@@ -25,7 +25,7 @@ Package: clblast
 Maintainer: CLBlast Developers <CNugteren@users.noreply.github.com>
 Architecture: amd64
 Depends: libc6 (>= 2.14), ocl-icd-libopencl1 | libopencl1, ocl-icd-libopencl1 (>= 1.0) | libopencl-1.1-1
-Version: ${CLBLASTVERSION}-1+${PLATFORM}
+Version: ${CLBLASTVERSION}
 Homepage: https://github.com/CNugteren/CLBlast/
 Description: The tuned OpenCL BLAS library
  CLBlast is a modern, lightweight, performant and tunable OpenCL BLAS library
@@ -36,9 +36,10 @@ EOS
 mv ${TARGET}/usr/lib/pkgconfig/clblast.pc ./clblast.pc.orig
 cat ./clblast.pc.orig  | sed -e s/^prefix=.*$/prefix=\\/usr/ > ${TARGET}/usr/lib/pkgconfig/clblast.pc
 rm ./clblast.pc.orig
-rm clblast_${CLBLASTVERSION}-1+${PLATFORM}_amd64.deb
+rm clblast_${CLBLASTVERSION}_amd64.deb
 fakeroot dpkg-deb --build pkgwork .
-rm ${FILENAME}.zip
-rm ${FILENAME}.tar.gz
+rm -f ${FILENAME}.tar.xz
+rm -f ${FILENAME}.zip
+rm -f ${FILENAME}.tar.gz
 rm -rf ${FILENAME}
 rm -rf ${TARGET}

phpstan-interop.neon

@@ -0,0 +1,22 @@
+parameters:
+	ignoreErrors:
+		-
+			message: "#^Call to an undefined method Interop\\\\Polite\\\\Math\\\\Matrix\\\\.*[Device\\|Linear]*Buffer\\:\\:dtype\\(\\)#"
+		-
+			message: "#^Call to an undefined method Interop\\\\Polite\\\\Math\\\\Matrix\\\\.*[Device\\|Linear]*Buffer\\:\\:addr\\(\\)#"
+
+
+#		-
+#			message: "#^Call to an undefined method Interop\\\\Polite\\\\Math\\\\Matrix\\\\.*[Device\\|Linear]*Buffer\\:\\:dump\\(\\)#"
+#		-
+#			message: "#^Call to an undefined method Interop\\\\Polite\\\\Math\\\\Matrix\\\\.*[Device\\|Linear]*Buffer\\:\\:value_size\\(\\)#"
+#		-
+#			message: "#function count expects array|Countable, Interop\\\\Polite\\\\Math\\\\Matrix\\\\NDArray given#"
+#		-
+#			message: "#^Argument of an invalid type Interop\\\\Polite\\\\Math\\\\Matrix\\\\NDArray supplied for foreach, only iterables are supported#"
+#		-
+#			message: "#^Call to an undefined method Interop\\\\Polite\\\\Math\\\\Matrix\\\\DeviceBuffer\\:\\:bytes\\(\\)#"
+#		-
+#			message: "#^Call to an undefined method Interop\\\\Polite\\\\Math\\\\Matrix\\\\DeviceBuffer\\:\\:copy\\(\\)#"
+#		-
+#			message: "#^Call to an undefined method Interop\\\\Polite\\\\Math\\\\Matrix\\\\DeviceBuffer\\:\\:read\\(\\)#"

phpstan.neon.dist

@@ -0,0 +1,10 @@
+includes:
+	- phpstan-interop.neon
+
+parameters:
+	level: 6
+	paths:
+		- src
+	ignoreErrors:
+		-
+			message: "#^Call to an undefined method FFI::#"

src/Blas.php

@@ -597,6 +597,7 @@ public function copy(
                 try {
                     $Y->copy($queue,$X,$bytes,$src_offset,$dst_offset,$event);
                 } catch(RuntimeException $e) {
+                    $status = $e->getCode();
                     throw new RuntimeException("CLBlast?copy error=$status", $status, $e);
                 }
                 // skip to move raw event 
@@ -881,6 +882,145 @@ public function rot(
         }
     }
 
+    public function rotmg(
+        DeviceBuffer $D1, int $offsetD1,
+        DeviceBuffer $D2, int $offsetD2,
+        DeviceBuffer $B1, int $offsetB1,
+        DeviceBuffer $B2, int $offsetB2,
+        DeviceBuffer $P,  int $offsetP,
+        CommandQueue $queue,// Rindow\OpenCL\CommandQueue
+        EventList $event=null,   // Rindow\OpenCL\EventList
+        ) : void
+    {
+        $ffi= $this->ffi;
+
+        //// Check Buffer A
+        //$this->assert_vector_buffer_spec("A", $A, 1, $offsetA, 1);
+        //// Check Buffer B
+        //$this->assert_vector_buffer_spec("B", $B, 1, $offsetB, 1);
+        //// Check Buffer C
+        //$this->assert_vector_buffer_spec("C", $C, 1, $offsetC, 1);
+        //// Check Buffer S
+        //$this->assert_vector_buffer_spec("S", $S, 1, $offsetS, 1);
+
+        // Check Buffer A and B and C and S
+        $dtype = $D1->dtype();
+        if($dtype!=$D2->dtype()||$dtype!=$B1->dtype()
+            ||$dtype!=$B2->dtype()||$dtype!=$P->dtype()) {
+            throw new InvalidArgumentException("Unmatch data type for A,B,C and S");
+        }
+        $bufferD1_p = $ffi->cast("cl_mem",$D1->_getId());
+        $bufferD2_p = $ffi->cast("cl_mem",$D2->_getId());
+        $bufferB1_p = $ffi->cast("cl_mem",$B1->_getId());
+        $bufferB2_p = $ffi->cast("cl_mem",$B2->_getId());
+        $bufferP_p = $ffi->cast("cl_mem",$P->_getId());
+        $queue_p = $ffi->cast("cl_command_queue*",FFI::addr($queue->_getId()));
+        $event_p = null;
+        if($event) {
+            $event_obj = $event->_ffi()->new("cl_event[1]");
+            $event_p = $ffi->cast("cl_event[1]",$event_obj);
+        }
+
+        switch($dtype) {
+            case NDArray::float32:{
+                $status = $ffi->CLBlastSrotmg(
+                    $bufferD1_p,$offsetD1,
+                    $bufferD2_p,$offsetD2,
+                    $bufferB1_p,$offsetB1,
+                    $bufferB2_p,$offsetB2,
+                    $bufferP_p,$offsetP,
+                    $queue_p,$event_p
+                );
+                break;
+            }
+            case NDArray::float64:{
+                $status = $ffi->CLBlastDrotmg(
+                    $bufferD1_p,$offsetD1,
+                    $bufferD2_p,$offsetD2,
+                    $bufferB1_p,$offsetB1,
+                    $bufferB2_p,$offsetB2,
+                    $bufferP_p,$offsetP,
+                    $queue_p,$event_p
+                );
+                break;
+            }
+            default: {
+                throw new InvalidArgumentException('Unsuppored data type');
+            }
+        }
+        if($status!=self::CLBlastSuccess) {
+            if($status==self::CLBlastNotImplemented) {
+                throw new RuntimeException("CLBlast?rotmg error=$status: Not Implemented", $status);
+            }
+            throw new RuntimeException("CLBlast?rotmg error=$status", $status);
+        }
+        if($event) {
+            $event->_move($event_obj);
+        }
+    }
+
+    public function rotm(
+        int $n,
+        DeviceBuffer $X, int $offsetX, int $incX,
+        DeviceBuffer $Y, int $offsetY, int $incY,
+        DeviceBuffer $P, int $offsetP,
+        CommandQueue $queue,// Rindow\OpenCL\CommandQueue
+        EventList $event=null,   // Rindow\OpenCL\EventList
+        ) : void
+    {
+        $ffi= $this->ffi;
+
+        // Check Buffer A and B and C and S
+        $dtype = $X->dtype();
+        if($dtype!=$Y->dtype()) {
+            throw new InvalidArgumentException("Unmatch data type for X and Y");
+        }
+        $bufferX_p = $ffi->cast("cl_mem",$X->_getId());
+        $bufferY_p = $ffi->cast("cl_mem",$Y->_getId());
+        $bufferP_p = $ffi->cast("cl_mem",$P->_getId());
+        $queue_p = $ffi->cast("cl_command_queue*",FFI::addr($queue->_getId()));
+        $event_p = null;
+        if($event) {
+            $event_obj = $event->_ffi()->new("cl_event[1]");
+            $event_p = $ffi->cast("cl_event[1]",$event_obj);
+        }
+
+        switch($dtype) {
+            case NDArray::float32:{
+                $status = $ffi->CLBlastSrotm(
+                    $n,
+                    $bufferX_p,$offsetX,$incX,
+                    $bufferY_p,$offsetY,$incY,
+                    $bufferP_p,$offsetP,
+                    $queue_p,$event_p
+                );
+                break;
+            }
+            case NDArray::float64:{
+                $status = $ffi->CLBlastDrotm(
+                    $n,
+                    $bufferX_p,$offsetX,$incX,
+                    $bufferY_p,$offsetY,$incY,
+                    $bufferP_p,$offsetP,
+                    $queue_p,$event_p
+                );
+                break;
+            }
+            default: {
+                throw new InvalidArgumentException('Unsuppored data type');
+            }
+        }
+        if($status!=self::CLBlastSuccess) {
+            if($status==self::CLBlastNotImplemented) {
+                throw new RuntimeException("CLBlast?rotm error=$status: Not Implemented", $status);
+            }
+            throw new RuntimeException("CLBlast?rotm error=$status", $status);
+        }
+        if($event) {
+            $event->_move($event_obj);
+        }
+    }
+
     public function gemv(
         int $order,
         int $trans,

src/CLBlastFactory.php

@@ -10,9 +10,14 @@ class CLBlastFactory
 {
     private static ?FFI $ffi = null;
     public static ?FFI $ffipf = null;
+    /** @var array<string> $libs_win */
     protected array $libs_win = ['clblast.dll'];
-    protected array $libs_linux = ['libclblast.so'];
+    /** @var array<string> $libs_linux */
+    protected array $libs_linux = ['libclblast.so.1'];
 
+    /**
+     * @param array<string> $libFiles
+     */
     public function __construct(
         string $headerFile=null,
         array $libFiles=null,

src/Math.php

@@ -7,6 +7,7 @@
 use InvalidArgumentException;
 use OutOfRangeException;
 use LogicException;
+use RuntimeException;
 use FFI;
 use Rindow\OpenCL\FFI\Buffer as DeviceBuffer;
 use Rindow\OpenCL\FFI\CommandQueue;
@@ -455,7 +456,7 @@ public function axpyBatched(
         DeviceBuffer $X, HostBuffer $offsetsX, int $offsetX, int $incX,
         DeviceBuffer $Y, HostBuffer $offsetsY, int $offsetY, int $incY,
         int $batch_count,
-        CommandQueue $queue, EventList $event
+        CommandQueue $queue, EventList $event=null
     ) : void
     {
         $ffi = $this->ffi;

src/Utils.php

@@ -2,6 +2,12 @@
 namespace Rindow\CLBlast\FFI;
 
 use Interop\Polite\Math\Matrix\NDArray;
+use InvalidArgumentException;
+
+class cl_float2_t {
+    /** @var array<float> $s */
+    public array $s;
+}
 
 trait Utils
 {
@@ -10,12 +16,14 @@ protected function toComplex(object $from,int $dtype) : object
         $ffi = $this->ffi;
         switch($dtype) {
             case NDArray::complex64: {
+                /** @var cl_float2_t $to */
                 $to = $ffi->new('cl_float2');
                 $to->s[0] = $from->real;
                 $to->s[1] = $from->imag;
                 break;
             }
             case NDArray::complex128: {
+                /** @var cl_float2_t $to */
                 $to = $ffi->new('cl_double2');
                 $to->s[0] = $from->real;
                 $to->s[1] = $from->imag;