Merge pull request #1055 from aroulin/shader-sge-sgei-slt

Shader: Implement SGE, SGEI and SLT in interpreter/JIT

3 files changed, 50 insertions, 15 deletions

diff --git a/src/video_core/shader/shader_interpreter.cpp b/src/video_core/shader/shader_interpreter.cpp
index 646171a19..063cc38f0 100644
--- a/src/video_core/shader/shader_interpreter.cpp
+++ b/src/video_core/shader/shader_interpreter.cpp
@@ -278,6 +278,20 @@ void RunInterpreter(UnitState<Debug>& state) {
                 break;
             }
 
+            case OpCode::Id::SGE:
+            case OpCode::Id::SGEI:
+                Record<DebugDataRecord::SRC1>(state.debug, iteration, src1);
+                Record<DebugDataRecord::SRC2>(state.debug, iteration, src2);
+                Record<DebugDataRecord::DEST_IN>(state.debug, iteration, dest);
+                for (int i = 0; i < 4; ++i) {
+                    if (!swizzle.DestComponentEnabled(i))
+                        continue;
+
+                    dest[i] = (src1[i] >= src2[i]) ? float24::FromFloat32(1.0f) : float24::FromFloat32(0.0f);
+                }
+                Record<DebugDataRecord::DEST_OUT>(state.debug, iteration, dest);
+                break;
+
             case OpCode::Id::SLT:
             case OpCode::Id::SLTI:
                 Record<DebugDataRecord::SRC1>(state.debug, iteration, src1);
diff --git a/src/video_core/shader/shader_jit_x64.cpp b/src/video_core/shader/shader_jit_x64.cpp
index e4b8295b3..a1bdd8456 100644
--- a/src/video_core/shader/shader_jit_x64.cpp
+++ b/src/video_core/shader/shader_jit_x64.cpp
@@ -29,8 +29,8 @@ const JitFunction instr_table[64] = {
     &JitCompiler::Compile_LG2,      // lg2
     nullptr,                        // unknown
     &JitCompiler::Compile_MUL,      // mul
-    nullptr,                        // lge
-    nullptr,                        // slt
+    &JitCompiler::Compile_SGE,      // sge
+    &JitCompiler::Compile_SLT,      // slt
     &JitCompiler::Compile_FLR,      // flr
     &JitCompiler::Compile_MAX,      // max
     &JitCompiler::Compile_MIN,      // min
@@ -46,8 +46,8 @@ const JitFunction instr_table[64] = {
     nullptr,                        // unknown
     nullptr,                        // dphi
     nullptr,                        // unknown
-    nullptr,                        // sgei
-    &JitCompiler::Compile_SLTI,     // slti
+    &JitCompiler::Compile_SGE,      // sgei
+    &JitCompiler::Compile_SLT,      // slti
     nullptr,                        // unknown
     nullptr,                        // unknown
     nullptr,                        // unknown
@@ -386,6 +386,36 @@ void JitCompiler::Compile_MUL(Instruction instr) {
     Compile_DestEnable(instr, SRC1);
 }
 
+void JitCompiler::Compile_SGE(Instruction instr) {
+    if (instr.opcode.Value().EffectiveOpCode() == OpCode::Id::SGEI) {
+        Compile_SwizzleSrc(instr, 1, instr.common.src1i, SRC1);
+        Compile_SwizzleSrc(instr, 2, instr.common.src2i, SRC2);
+    } else {
+        Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
+        Compile_SwizzleSrc(instr, 2, instr.common.src2, SRC2);
+    }
+
+    CMPPS(SRC1, R(SRC2), CMP_NLT);
+    ANDPS(SRC1, R(ONE));
+
+    Compile_DestEnable(instr, SRC1);
+}
+
+void JitCompiler::Compile_SLT(Instruction instr) {
+    if (instr.opcode.Value().EffectiveOpCode() == OpCode::Id::SLTI) {
+        Compile_SwizzleSrc(instr, 1, instr.common.src1i, SRC1);
+        Compile_SwizzleSrc(instr, 2, instr.common.src2i, SRC2);
+    } else {
+        Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
+        Compile_SwizzleSrc(instr, 2, instr.common.src2, SRC2);
+    }
+
+    CMPPS(SRC1, R(SRC2), CMP_LT);
+    ANDPS(SRC1, R(ONE));
+
+    Compile_DestEnable(instr, SRC1);
+}
+
 void JitCompiler::Compile_FLR(Instruction instr) {
     Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
 
@@ -463,16 +493,6 @@ void JitCompiler::Compile_MOV(Instruction instr) {
     Compile_DestEnable(instr, SRC1);
 }
 
-void JitCompiler::Compile_SLTI(Instruction instr) {
-    Compile_SwizzleSrc(instr, 1, instr.common.src1i, SRC1);
-    Compile_SwizzleSrc(instr, 1, instr.common.src2i, SRC2);
-
-    CMPSS(SRC1, R(SRC2), CMP_LT);
-    ANDPS(SRC1, R(ONE));
-
-    Compile_DestEnable(instr, SRC1);
-}
-
 void JitCompiler::Compile_RCP(Instruction instr) {
     Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
 
diff --git a/src/video_core/shader/shader_jit_x64.h b/src/video_core/shader/shader_jit_x64.h
index a6ae7fbf1..b2aa5293c 100644
--- a/src/video_core/shader/shader_jit_x64.h
+++ b/src/video_core/shader/shader_jit_x64.h
@@ -40,6 +40,8 @@ public:
     void Compile_EX2(Instruction instr);
     void Compile_LG2(Instruction instr);
     void Compile_MUL(Instruction instr);
+    void Compile_SGE(Instruction instr);
+    void Compile_SLT(Instruction instr);
     void Compile_FLR(Instruction instr);
     void Compile_MAX(Instruction instr);
     void Compile_MIN(Instruction instr);
@@ -47,7 +49,6 @@ public:
     void Compile_RSQ(Instruction instr);
     void Compile_MOVA(Instruction instr);
     void Compile_MOV(Instruction instr);
-    void Compile_SLTI(Instruction instr);
     void Compile_NOP(Instruction instr);
     void Compile_END(Instruction instr);
     void Compile_CALL(Instruction instr);