[IGC feature][staged compilation] Add fastest stage 1 to emit SIMD8 w/o opts

Change-Id: I64d5a3b7af5f4c13441eb6994961f3b9535d787e

Author: houjenko

IGC/AdaptorCommon/API/igc.h

@@ -108,12 +108,14 @@ typedef enum {
     FLAG_CG_ALL_SIMDS = 0,
     FLAG_CG_STAGE1_FAST_COMPILE = 1,
     FLAG_CG_STAGE1_BEST_PERF = 2,
+    FLAG_CG_STAGE1_FASTEST_COMPILE = 3,
 } CG_FLAG_t;
 
 #define IsStage2Available(ctx_ptr) (ctx_ptr != nullptr)
 
 #define IsStage2RestSIMDs(prev_ctx_ptr) (prev_ctx_ptr != nullptr)
 #define IsStage1FastCompile(flag, prev_ctx_ptr) (!IsStage2RestSIMDs(prev_ctx_ptr) && flag == FLAG_CG_STAGE1_FAST_COMPILE)
+#define IsStage1FastestCompile(flag, prev_ctx_ptr) (!IsStage2RestSIMDs(prev_ctx_ptr) && flag == FLAG_CG_STAGE1_FASTEST_COMPILE)
 #define IsStage1BestPerf(flag, prev_ctx_ptr)    (!IsStage2RestSIMDs(prev_ctx_ptr) && flag == FLAG_CG_STAGE1_BEST_PERF)
 #define IsAllSIMDs(flag, prev_ctx_ptr)          (!IsStage2RestSIMDs(prev_ctx_ptr) && flag == FLAG_CG_ALL_SIMDS)
 #define IsStage1(pCtx)   (IsStage1BestPerf(pCtx->m_CgFlag, pCtx->m_StagingCtx) || \

IGC/Compiler/CISACodeGen/CISABuilder.cpp

@@ -4383,6 +4383,14 @@ namespace IGC
 
         // Set to stitch all functions to all kernels in a VISABuidler
         SaveOption(vISA_noStitchExternFunc, false);
+
+        // Turning off optimizations as much as possible to have the fastest compilation
+        if (IsStage1FastestCompile(context->m_CgFlag, context->m_StagingCtx))
+        {
+            SaveOption(vISA_FastSpill, true);
+            SaveOption(vISA_LocalScheduling, false);
+            SaveOption(vISA_preRA_Schedule, false);
+        }
     }
 
     void CEncoder::InitEncoder(bool canAbortOnSpill, bool hasStackCall)

IGC/Compiler/CISACodeGen/ShaderCodeGen.cpp

@@ -844,7 +844,8 @@ static void PSCodeGen(
         AddCodeGenPasses(*ctx, shaders, PassMgr, SIMDMode::SIMD8, !ctx->m_retryManager.IsLastTry(), ShaderDispatchMode::NOT_APPLICABLE, pSignature);
         useRegKeySimd = true;
     }
-    else if (IsStage1FastCompile(ctx->m_CgFlag, ctx->m_StagingCtx))
+    else if (IsStage1FastCompile(ctx->m_CgFlag, ctx->m_StagingCtx) ||
+             IsStage1FastestCompile(ctx->m_CgFlag, ctx->m_StagingCtx))
     {
         AddCodeGenPasses(*ctx, shaders, PassMgr, SIMDMode::SIMD8, false, ShaderDispatchMode::NOT_APPLICABLE, pSignature);
         useRegKeySimd = true;

IGC/common/debug/Dump.cpp

@@ -330,11 +330,15 @@ std::string DumpName::AbsolutePath(OutputFolderName folder) const
         {
             ss << "FastStage1";
         }
-        else
+        else if (m_cgFlag.getValue() == FLAG_CG_STAGE1_BEST_PERF)
         {
-            IGC_ASSERT(m_cgFlag.getValue() == FLAG_CG_STAGE1_BEST_PERF);
             ss << "BestStage1";
         }
+        else
+        {
+            IGC_ASSERT(m_cgFlag.getValue() == FLAG_CG_STAGE1_FASTEST_COMPILE);
+            ss << "FastestStage1";
+        }
 
         underscore = true;
     }

IGC/common/igc_flags.def

@@ -381,7 +381,7 @@ DECLARE_IGC_REGKEY(bool, StagedCompilation,             false, "Enable staged co
 DECLARE_IGC_REGKEY(bool, EnableTrackPtr,                false, "Track Staging Context alloc/dealloc", false)
 DECLARE_IGC_REGKEY(bool, ExtraRetrySIMD16,              false,  "Enable extra simd16 with retry for STAGE1_BEST_PREF", false)
 DECLARE_IGC_REGKEY(bool, SaveRestoreIR,                 true,  "Save/Restore IR for staged compilation to avoid duplicated compilations", false)
-DECLARE_IGC_REGKEY(bool, StagedBestSIMD, false,  "Force Pixel shader to return the best SIMD, either SIMD16 or SIMD8. When StagedCompilation is on, this only affects to the first stage", false)
+DECLARE_IGC_REGKEY(DWORD, FirstStagedSIMD,              1,      "Force Pixel shader to be 1: BestSIMD (SIMD16 or SIMD8), 2: FastSIMD (SIMD8), 3: FatestSIMD (SIMD8 opt off)", false)
 DECLARE_IGC_REGKEY(bool, ForceBestSIMD, false,  "Force pixel shader to return the best SIMD, either SIMD16 or SIMD8.", false)
 DECLARE_IGC_REGKEY(bool, SkipTREarlyExitCheck, false, "Skip SIMD16 early exit check in ShaderCodeGen", false)
 DECLARE_IGC_REGKEY(bool, EnableTCSHWBarriers, false,  "Enable TCS pass with HW barriers support. Default TCS pass is TCS pass with multiple continuation functions.", false)