Fixed CUDA warning and restored Matrix5d inversion on GPU.

ericcano · ericcano · commit daa09a43686b · 2021-08-03T16:51:04.000+02:00
diff --git a/Eigen/src/Core/products/GeneralBlockPanelKernel.h b/Eigen/src/Core/products/GeneralBlockPanelKernel.h
@@ -1312,6 +1312,7 @@ struct lhs_process_one_packet
 {
   typedef typename GEBPTraits::RhsPacketx4 RhsPacketx4;
 
+  EIGEN_DEVICE_FUNC
   EIGEN_STRONG_INLINE void peeled_kc_onestep(Index K, const LhsScalar* blA, const RhsScalar* blB, GEBPTraits traits, LhsPacket *A0, RhsPacketx4 *rhs_panel, RhsPacket *T0, AccPacket *C0, AccPacket *C1, AccPacket *C2, AccPacket *C3)
   {
     EIGEN_ASM_COMMENT("begin step of gebp micro kernel 1X4");
@@ -1328,6 +1329,7 @@ struct lhs_process_one_packet
     EIGEN_ASM_COMMENT("end step of gebp micro kernel 1X4");
   }
 
+  EIGEN_DEVICE_FUNC
   EIGEN_STRONG_INLINE void operator()(
     const DataMapper& res, const LhsScalar* blockA, const RhsScalar* blockB, ResScalar alpha,
     Index peelStart, Index peelEnd, Index strideA, Index strideB, Index offsetA, Index offsetB,

Original file line number	Diff line number	Diff line change
`@@ -1312,6 +1312,7 @@ struct lhs_process_one_packet`
`1312`	`1312`	`{`
`1313`	`1313`	`typedef typename GEBPTraits::RhsPacketx4 RhsPacketx4;`
`1314`	`1314`
	`1315`	`+ EIGEN_DEVICE_FUNC`
`1315`	`1316`	`EIGEN_STRONG_INLINE void peeled_kc_onestep(Index K, const LhsScalar* blA, const RhsScalar* blB, GEBPTraits traits, LhsPacket A0, RhsPacketx4 rhs_panel, RhsPacket T0, AccPacket C0, AccPacket C1, AccPacket C2, AccPacket *C3)`
`1316`	`1317`	`{`
`1317`	`1318`	`EIGEN_ASM_COMMENT("begin step of gebp micro kernel 1X4");`
`@@ -1328,6 +1329,7 @@ struct lhs_process_one_packet`
`1328`	`1329`	`EIGEN_ASM_COMMENT("end step of gebp micro kernel 1X4");`
`1329`	`1330`	`}`
`1330`	`1331`
	`1332`	`+ EIGEN_DEVICE_FUNC`
`1331`	`1333`	`EIGEN_STRONG_INLINE void operator()(`
`1332`	`1334`	`const DataMapper& res, const LhsScalar* blockA, const RhsScalar* blockB, ResScalar alpha,`
`1333`	`1335`	`Index peelStart, Index peelEnd, Index strideA, Index strideB, Index offsetA, Index offsetB,`