PPC [wasm-simd]: Implement saturating rounding multiply high

Bug: v8:10971
Change-Id: Idaa75b5c4d63695dbb8eed2be076f067ff5df9ff
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2623817Reviewed-by: Junliang Yan <junyan@redhat.com>
Commit-Queue: Milad Fa <mfarazma@redhat.com>
Cr-Commit-Position: refs/heads/master@{#72050}

src/codegen/ppc/constants-ppc.h

@@ -1936,7 +1936,9 @@ using Instr = uint32_t;
   /* Vector Select */                                           \
   V(vsel, VSEL, 0x1000002A)                                     \
   /* Vector Multiply-Sum Signed Halfword Modulo */              \
-  V(vmsumshm, VMSUMSHM, 0x10000028)
+  V(vmsumshm, VMSUMSHM, 0x10000028)                             \
+  /* Vector Multiply-High-Round-Add Signed Halfword Saturate */ \
+  V(vmhraddshs, VMHRADDSHS, 0x10000021)
 
 #define PPC_VA_OPCODE_UNUSED_LIST(V)                             \
   /* Vector Add Extended & write Carry Unsigned Quadword */      \
@@ -1947,8 +1949,6 @@ using Instr = uint32_t;
   V(vmaddfp, VMADDFP, 0x1000002E)                                \
   /* Vector Multiply-High-Add Signed Halfword Saturate */        \
   V(vmhaddshs, VMHADDSHS, 0x10000020)                            \
-  /* Vector Multiply-High-Round-Add Signed Halfword Saturate */  \
-  V(vmhraddshs, VMHRADDSHS, 0x10000021)                          \
   /* Vector Multiply-Sum Mixed Byte Modulo */                    \
   V(vmsummbm, VMSUMMBM, 0x10000025)                              \
   /* Vector Multiply-Sum Signed Halfword Saturate */             \

src/compiler/backend/ppc/code-generator-ppc.cc

@@ -3714,6 +3714,12 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       break;
     }
 #undef EXT_ADD_PAIRWISE
+    case kPPC_I16x8Q15MulRSatS: {
+      __ vxor(kScratchSimd128Reg, kScratchSimd128Reg, kScratchSimd128Reg);
+      __ vmhraddshs(i.OutputSimd128Register(), i.InputSimd128Register(0),
+                    i.InputSimd128Register(1), kScratchSimd128Reg);
+      break;
+    }
     case kPPC_StoreCompressTagged: {
       ASSEMBLE_STORE_INTEGER(StoreTaggedField, StoreTaggedFieldX);
       break;

src/compiler/backend/ppc/instruction-codes-ppc.h

@@ -335,6 +335,7 @@ namespace compiler {
   V(PPC_I16x8BitMask)                \
   V(PPC_I16x8ExtAddPairwiseI8x16S)   \
   V(PPC_I16x8ExtAddPairwiseI8x16U)   \
+  V(PPC_I16x8Q15MulRSatS)            \
   V(PPC_I8x16Splat)                  \
   V(PPC_I8x16ExtractLaneU)           \
   V(PPC_I8x16ExtractLaneS)           \

src/compiler/backend/ppc/instruction-scheduler-ppc.cc

@@ -258,6 +258,7 @@ int InstructionScheduler::GetTargetInstructionFlags(
     case kPPC_I16x8BitMask:
     case kPPC_I16x8ExtAddPairwiseI8x16S:
     case kPPC_I16x8ExtAddPairwiseI8x16U:
+    case kPPC_I16x8Q15MulRSatS:
     case kPPC_I8x16Splat:
     case kPPC_I8x16ExtractLaneU:
     case kPPC_I8x16ExtractLaneS:

src/compiler/backend/ppc/instruction-selector-ppc.cc

@@ -2210,6 +2210,7 @@ void InstructionSelector::VisitInt64AbsWithOverflow(Node* node) {
   V(I16x8AddSatU)          \
   V(I16x8SubSatU)          \
   V(I16x8RoundingAverageU) \
+  V(I16x8Q15MulRSatS)      \
   V(I8x16Add)              \
   V(I8x16Sub)              \
   V(I8x16Mul)              \