Changeset 103768 in vbox for trunk/src/VBox/VMM/VMMAll

Timestamp:

Mar 11, 2024 2:39:16 PM (9 months ago)

Author:

vboxsync

Message:

VMM/IEM: Some more threaded function name annotating to make the native recompilation state more visible in the statistics. bugref:10376

Location:

trunk/src/VBox/VMM/VMMAll

Files:

• IEMAllN8vePython.py (modified) (2 diffs)
• IEMAllThrdPython.py (modified) (7 diffs)

trunk/src/VBox/VMM/VMMAll/IEMAllN8vePython.py

@@ -659 +659 @@
 
 
-def displayStatistics(aoThreadedFuncs, sHostArch): # type (list(ThreadedFunction)) -> True
+def analyzeThreadedFunctionsForNativeRecomp(aoThreadedFuncs, sHostArch): # type (list(ThreadedFunction)) -> True
     """
     Displays statistics.
@@ -667 +667 @@
     cNative = 0;
     for oThreadedFunction in aoThreadedFuncs:
+        cNativeVariations = 0;
         for oVariation in oThreadedFunction.aoVariations:
             cTotal += 1;
             oVariation.oNativeRecomp = analyzeVariantForNativeRecomp(oVariation, sHostArch);
             if oVariation.oNativeRecomp and oVariation.oNativeRecomp.isRecompilable():
-                cNative += 1;
+                cNativeVariations += 1;
+        cNative += cNativeVariations;
+
+        # If all variations can be recompiled natively, annotate the threaded
+        # function name accordingly so it'll be easy to spot in the stats.
+        if oThreadedFunction.sSubName:
+            if cNativeVariations == len(oThreadedFunction.aoVariations):
+                aoStmts = oThreadedFunction.oMcBlock.decode();
+                oStmt = iai.McStmt.findStmtByNames(aoStmts, {'IEM_MC_NATIVE_IF': True,});
+                if oStmt and NativeRecompFunctionVariation.kdOptionArchToVal[sHostArch] in oStmt.asArchitectures:
+                    oThreadedFunction.sSubName += '_ne';        # native emit
+                elif oThreadedFunction.sSubName.find('aimpl') >= 0:
+                    oThreadedFunction.sSubName += '_na';        # native aimpl
+                else:
+                    oThreadedFunction.sSubName += '_nn';        # native native
+            elif cNativeVariations == 0:
+                oThreadedFunction.sSubName += '_ntodo';         # native threaded todo
+            else:
+                oThreadedFunction.sSubName += '_nm';            # native mixed
+
+
     print('todo: %.1f%% / %u out of %u threaded function variations are recompilable'
           % (cNative * 100.0 / cTotal, cNative, cTotal), file = sys.stderr);

trunk/src/VBox/VMM/VMMAll/IEMAllThrdPython.py

@@ -1564 +1564 @@
         self.dsCImplFlags   = {}            # type: Dict[str, bool]
         ## The unique sub-name for this threaded function.
-        self.sSubName  = '';
+        self.sSubName       = '';
         #if oMcBlock.iInFunction > 0 or (oMcBlock.oInstruction and len(oMcBlock.oInstruction.aoMcBlocks) > 1):
         #    self.sSubName  = '_%s' % (oMcBlock.iInFunction);
@@ -1835 +1835 @@
         'IEM_MC_MERGE_YREG_U64LOCAL_U64HI_ZX_VLMAX':'__yreg64',
     };
+    kdAnnotateNameCallStmts = {
+        'IEM_MC_CALL_CIMPL_0':                      '__cimpl',
+        'IEM_MC_CALL_CIMPL_1':                      '__cimpl',
+        'IEM_MC_CALL_CIMPL_2':                      '__cimpl',
+        'IEM_MC_CALL_CIMPL_3':                      '__cimpl',
+        'IEM_MC_CALL_CIMPL_4':                      '__cimpl',
+        'IEM_MC_CALL_CIMPL_5':                      '__cimpl',
+        'IEM_MC_CALL_CIMPL_6':                      '__cimpl',
+        'IEM_MC_CALL_CIMPL_7':                      '__cimpl',
+        'IEM_MC_DEFER_TO_CIMPL_0_RET':              '__cimpl_defer',
+        'IEM_MC_DEFER_TO_CIMPL_1_RET':              '__cimpl_defer',
+        'IEM_MC_DEFER_TO_CIMPL_2_RET':              '__cimpl_defer',
+        'IEM_MC_DEFER_TO_CIMPL_3_RET':              '__cimpl_defer',
+        'IEM_MC_DEFER_TO_CIMPL_4_RET':              '__cimpl_defer',
+        'IEM_MC_DEFER_TO_CIMPL_5_RET':              '__cimpl_defer',
+        'IEM_MC_DEFER_TO_CIMPL_6_RET':              '__cimpl_defer',
+        'IEM_MC_DEFER_TO_CIMPL_7_RET':              '__cimpl_defer',
+        'IEM_MC_CALL_VOID_AIMPL_0':                 '__aimpl',
+        'IEM_MC_CALL_VOID_AIMPL_1':                 '__aimpl',
+        'IEM_MC_CALL_VOID_AIMPL_2':                 '__aimpl',
+        'IEM_MC_CALL_VOID_AIMPL_3':                 '__aimpl',
+        'IEM_MC_CALL_VOID_AIMPL_4':                 '__aimpl',
+        'IEM_MC_CALL_VOID_AIMPL_5':                 '__aimpl',
+        'IEM_MC_CALL_AIMPL_0':                      '__aimpl_ret',
+        'IEM_MC_CALL_AIMPL_1':                      '__aimpl_ret',
+        'IEM_MC_CALL_AIMPL_2':                      '__aimpl_ret',
+        'IEM_MC_CALL_AIMPL_3':                      '__aimpl_ret',
+        'IEM_MC_CALL_AIMPL_4':                      '__aimpl_ret',
+        'IEM_MC_CALL_AIMPL_5':                      '__aimpl_ret',
+        'IEM_MC_CALL_AIMPL_6':                      '__aimpl_ret',
+        'IEM_MC_CALL_VOID_AIMPL_6':                 '__aimpl_fpu',
+        'IEM_MC_CALL_FPU_AIMPL_0':                  '__aimpl_fpu',
+        'IEM_MC_CALL_FPU_AIMPL_1':                  '__aimpl_fpu',
+        'IEM_MC_CALL_FPU_AIMPL_2':                  '__aimpl_fpu',
+        'IEM_MC_CALL_FPU_AIMPL_3':                  '__aimpl_fpu',
+        'IEM_MC_CALL_FPU_AIMPL_4':                  '__aimpl_fpu',
+        'IEM_MC_CALL_FPU_AIMPL_5':                  '__aimpl_fpu',
+        'IEM_MC_CALL_MMX_AIMPL_0':                  '__aimpl_mmx',
+        'IEM_MC_CALL_MMX_AIMPL_1':                  '__aimpl_mmx',
+        'IEM_MC_CALL_MMX_AIMPL_2':                  '__aimpl_mmx',
+        'IEM_MC_CALL_MMX_AIMPL_3':                  '__aimpl_mmx',
+        'IEM_MC_CALL_MMX_AIMPL_4':                  '__aimpl_mmx',
+        'IEM_MC_CALL_MMX_AIMPL_5':                  '__aimpl_mmx',
+        'IEM_MC_CALL_SSE_AIMPL_0':                  '__aimpl_sse',
+        'IEM_MC_CALL_SSE_AIMPL_1':                  '__aimpl_sse',
+        'IEM_MC_CALL_SSE_AIMPL_2':                  '__aimpl_sse',
+        'IEM_MC_CALL_SSE_AIMPL_3':                  '__aimpl_sse',
+        'IEM_MC_CALL_SSE_AIMPL_4':                  '__aimpl_sse',
+        'IEM_MC_CALL_SSE_AIMPL_5':                  '__aimpl_sse',
+        'IEM_MC_CALL_AVX_AIMPL_0':                  '__aimpl_avx',
+        'IEM_MC_CALL_AVX_AIMPL_1':                  '__aimpl_avx',
+        'IEM_MC_CALL_AVX_AIMPL_2':                  '__aimpl_avx',
+        'IEM_MC_CALL_AVX_AIMPL_3':                  '__aimpl_avx',
+        'IEM_MC_CALL_AVX_AIMPL_4':                  '__aimpl_avx',
+        'IEM_MC_CALL_AVX_AIMPL_5':                  '__aimpl_avx',
+    };
     def analyzeAndAnnotateName(self, aoStmts: List[iai.McStmt]):
         """
@@ -1840 +1896 @@
         and sets self.sSubName if successfull.
         """
+        # Operand base naming:
         dHits = {};
         cHits = iai.McStmt.countStmtsByName(aoStmts, self.kdAnnotateNameMemStmts, dHits);
@@ -1847 +1904 @@
         else:
             cHits = iai.McStmt.countStmtsByName(aoStmts, self.kdAnnotateNameRegStmts, dHits);
-            if not cHits:
+            if cHits > 0:
+                sStmtNm = sorted(dHits.keys())[-1]; # priority: STORE, MEM_MAP, FETCH.
+                sName = self.kdAnnotateNameRegStmts[sStmtNm];
+            else:
+                # No op details, try name it by call type...
+                cHits = iai.McStmt.countStmtsByName(aoStmts, self.kdAnnotateNameCallStmts, dHits);
+                if cHits > 0:
+                    sStmtNm = sorted(dHits.keys())[-1]; # Not really necessary to sort, but simple this way.
+                    self.sSubName = self.kdAnnotateNameCallStmts[sStmtNm];
                 return;
-            sStmtNm = sorted(dHits.keys())[-1]; # priority: STORE, MEM_MAP, FETCH.
-            sName = self.kdAnnotateNameRegStmts[sStmtNm];
-
-        oStmt = iai.McStmt.findStmtByNames(aoStmts, {'IEM_MC_NATIVE_IF': True,});
-        if oStmt and oStmt.asArchitectures:
-            sName += '_n'; ## @todo check if enabled for the host architecture
+
+        # Add call info if any:
+        dHits = {};
+        cHits = iai.McStmt.countStmtsByName(aoStmts, self.kdAnnotateNameCallStmts, dHits);
+        if cHits > 0:
+            sStmtNm = sorted(dHits.keys())[-1]; # Not really necessary to sort, but simple this way.
+            sName += self.kdAnnotateNameCallStmts[sStmtNm][1:];
+
         self.sSubName = sName;
         return;
@@ -1949 +2016 @@
         return sAnnotation;
 
-    def analyze(self, oGenerator):
+    def analyzeThreadedFunction(self, oGenerator):
         """
         Analyzes the code, identifying the number of parameters it requires and such.
@@ -2542 +2609 @@
         dMinParamCounts = {};
         for oThreadedFunction in self.aoThreadedFuncs:
-            oThreadedFunction.analyze(self);
+            oThreadedFunction.analyzeThreadedFunction(self);
             for oVariation in oThreadedFunction.aoVariations:
                 dRawParamCounts[len(oVariation.dParamRefs)] = dRawParamCounts.get(len(oVariation.dParamRefs), 0) + 1;
@@ -2584 +2651 @@
         # Analyze the threaded functions and their variations for native recompilation.
         if fNativeRecompilerEnabled:
-            ian.displayStatistics(self.aoThreadedFuncs, sHostArch);
+            ian.analyzeThreadedFunctionsForNativeRecomp(self.aoThreadedFuncs, sHostArch);
 
         # Gather arguments + variable statistics for the MC blocks.