summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authoraroulin <andy.roulin@epfl.ch>2015-08-19 01:49:45 +0200
committeraroulin <andy.roulin@epfl.ch>2015-08-19 03:40:07 +0200
commit2f9eb98f03b99482d3b79c7afb345d8c5014ddf8 (patch)
tree98f63e0a2ae6d44a85733e12605e1b49d45f1264
parentShader: implement EX2 and LG2 in JIT (diff)
downloadyuzu-2f9eb98f03b99482d3b79c7afb345d8c5014ddf8.tar
yuzu-2f9eb98f03b99482d3b79c7afb345d8c5014ddf8.tar.gz
yuzu-2f9eb98f03b99482d3b79c7afb345d8c5014ddf8.tar.bz2
yuzu-2f9eb98f03b99482d3b79c7afb345d8c5014ddf8.tar.lz
yuzu-2f9eb98f03b99482d3b79c7afb345d8c5014ddf8.tar.xz
yuzu-2f9eb98f03b99482d3b79c7afb345d8c5014ddf8.tar.zst
yuzu-2f9eb98f03b99482d3b79c7afb345d8c5014ddf8.zip
-rw-r--r--src/video_core/shader/shader_jit_x64.cpp30
-rw-r--r--src/video_core/shader/shader_jit_x64.h3
2 files changed, 33 insertions, 0 deletions
diff --git a/src/video_core/shader/shader_jit_x64.cpp b/src/video_core/shader/shader_jit_x64.cpp
index 93f608584..e4b8295b3 100644
--- a/src/video_core/shader/shader_jit_x64.cpp
+++ b/src/video_core/shader/shader_jit_x64.cpp
@@ -280,6 +280,22 @@ void JitCompiler::Compile_UniformCondition(Instruction instr) {
CMP(sizeof(bool) * 8, MDisp(UNIFORMS, offset), Imm8(0));
}
+void JitCompiler::Compile_PushCallerSavedXMM() {
+#ifndef _WIN32
+ SUB(64, R(RSP), Imm8(2 * 16));
+ MOVUPS(MDisp(RSP, 16), ONE);
+ MOVUPS(MDisp(RSP, 0), NEGBIT);
+#endif
+}
+
+void JitCompiler::Compile_PopCallerSavedXMM() {
+#ifndef _WIN32
+ MOVUPS(NEGBIT, MDisp(RSP, 0));
+ MOVUPS(ONE, MDisp(RSP, 16));
+ ADD(64, R(RSP), Imm8(2 * 16));
+#endif
+}
+
void JitCompiler::Compile_ADD(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2, SRC2);
@@ -334,7 +350,14 @@ void JitCompiler::Compile_DP4(Instruction instr) {
void JitCompiler::Compile_EX2(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
MOVSS(XMM0, R(SRC1));
+
+ // The following will actually break the stack alignment
+ ABI_PushAllCallerSavedRegsAndAdjustStack();
+ Compile_PushCallerSavedXMM();
ABI_CallFunction(reinterpret_cast<const void*>(exp2f));
+ Compile_PopCallerSavedXMM();
+ ABI_PopAllCallerSavedRegsAndAdjustStack();
+
SHUFPS(XMM0, R(XMM0), _MM_SHUFFLE(0, 0, 0, 0));
MOVAPS(SRC1, R(XMM0));
Compile_DestEnable(instr, SRC1);
@@ -343,7 +366,14 @@ void JitCompiler::Compile_EX2(Instruction instr) {
void JitCompiler::Compile_LG2(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
MOVSS(XMM0, R(SRC1));
+
+ // The following will actually break the stack alignment
+ ABI_PushAllCallerSavedRegsAndAdjustStack();
+ Compile_PushCallerSavedXMM();
ABI_CallFunction(reinterpret_cast<const void*>(log2f));
+ Compile_PopCallerSavedXMM();
+ ABI_PopAllCallerSavedRegsAndAdjustStack();
+
SHUFPS(XMM0, R(XMM0), _MM_SHUFFLE(0, 0, 0, 0));
MOVAPS(SRC1, R(XMM0));
Compile_DestEnable(instr, SRC1);
diff --git a/src/video_core/shader/shader_jit_x64.h b/src/video_core/shader/shader_jit_x64.h
index 104f9f466..a6ae7fbf1 100644
--- a/src/video_core/shader/shader_jit_x64.h
+++ b/src/video_core/shader/shader_jit_x64.h
@@ -69,6 +69,9 @@ private:
void Compile_EvaluateCondition(Instruction instr);
void Compile_UniformCondition(Instruction instr);
+ void Compile_PushCallerSavedXMM();
+ void Compile_PopCallerSavedXMM();
+
/// Pointer to the variable that stores the current Pica code offset. Used to handle nested code blocks.
unsigned* offset_ptr = nullptr;