For byval parameter, PPC currently will generate loads in the `lower_call()` to loads the parameter to the registers. If there is store instruction which stores some value to the parameter in the same block, there will be load hit store issue. This patch tries to solve this. Implementing this in PPC Peephole is because: 1: `lower_call` in PPC instruction selection pass has several versions for different platforms, so in peephole pass can avoid adding duplicated codes for each ABI implementation. 2: The load elimination should not be limited to a single block(even though for load-hit-store, it mostly happens in same block), for example, if we can analysis domination and alias well, it should be possible to eliminate a load in a different block with the store. So in DAG based PPC instruction selection pass, we can not extend current implementation to a function level optimization.

When lowering calls on target like PPC, some stack loads will be generated for by value parameters. Node CALLSEQ_START prevents such loads from being combined. Suggested by @RolandF, this patch removes the unnecessary loads for the byval parameter by extending `ForwardStoreValueToDirectLoad`

For byval parameter, PPC currently will generate loads in the `lower_call()` to loads the parameter to the registers. If there is store instruction which stores some value to the parameter in the same blockWhen lowering calls on target like PPC, there will besome stack load hit store issue. This patch tries to solve this.s will be generated Implementing this in PPC Peephole is because:for by value parameters. Node CALLSEQ_START prevents such loads from 1: `lower_call` in PPC instruction selection pass has several versions for different platformsbeing combined. Suggested by @RolandF, so in peephole pass can avoid adding duplicated codethis patch removes the unnecessary loads for each ABI implementation.the 2: The load elimination should not be limited to a single block(even though for load-hit-store, it mostly happens in same block), for example, if we can analysis domination and alias well, it should be possible to eliminate a load in a different block with the store. So in DAG based PPC instruction selection pass,byval parameter by extending `ForwardStoreValueToDirectLoad` we can not extend current implementation to a function level optimization.