FPROUND - variable flag offset

doc/isa-ops.md

@@ -12,6 +12,7 @@ There are 31 unique instructions divided into 3 groups:
 
 ## Integer instructions
 There are 22 integer instructions. They are divided into 3 classes (MATH, DIV, SHIFT) with different B operand selection rules.
+
 |# opcodes|instruction|class|signed|A width|B width|C|C width|
 |-|-|-|-|-|-|-|-|
 |12|ADD_64|MATH|no|64|64|`A + B`|64|
@@ -55,7 +56,7 @@ The shift/rotate instructions use just the bottom 6 bits of the `B` operand (`im
 There are 5 floating point instructions. All floating point instructions are vector instructions that operate on two packed double precision floating point values.
 
 |# opcodes|instruction|C|
-|-|-|-|-|
+|-|-|-|
 |20|FPADD|`A + B`|
 |20|FPSUB|`A - B`|
 |22|FPMUL|`A * B`|

doc/isa.md

@@ -9,6 +9,7 @@ The encoding of each 128-bit instruction word is following:
 There are 256 opcodes, which are distributed between 3 groups of instructions. There are 31 distinct operations (each operation can be encoded using multiple opcodes - for example opcodes `0x00` to `0x0d` correspond to integer addition).
 
 **Table 1: Instruction groups**
+
 |group|# operations|# opcodes||
 |---------|-----------------|----|-|
 |integer (IA)|22|144|56.3%|
@@ -31,8 +32,8 @@ The `A.LOC.W` flag determines the address width when reading operand A from the
 
 **Table 3: Operand A read address width**
 
-|`A.LOC.W`|address width (W)
-|---------|-|-|
+|`A.LOC.W`|address width (W)|
+|---------|-|
 |0|15 bits (256 KiB)|
 |1-3|11 bits (16 KiB)|
 
@@ -125,8 +126,8 @@ The `C.LOC.W` flag determines the address width when writing operand C to the sc
 
 **Table 10: Operand C write address width**
 
-|`C.LOC.W`|address width (W)
-|---------|-|-|
+|`C.LOC.W`|address width (W)|
+|---------|-|
 |0|15 bits (256 KiB)|
 |1-3|11 bits (16 KiB)|
 

src/AssemblyGeneratorX86.cpp

@@ -466,7 +466,9 @@ namespace RandomX {
 	void AssemblyGeneratorX86::h_FPROUND(Instruction& instr, int i) {
 		genar(instr, i);
 		asmCode << "\tmov rcx, rax" << std::endl;
-		asmCode << "\tshl eax, 13" << std::endl;
+		int rotate = (13 - (instr.imm8 & 63)) & 63;
+		if (rotate != 0)
+			asmCode << "\trol rax, " << rotate << std::endl;
 		asmCode << "\tand eax, 24576" << std::endl;
 		asmCode << "\tor eax, 40896" << std::endl;
 		asmCode << "\tmov dword ptr [rsp - 8], eax" << std::endl;

src/JitCompilerX86.cpp

@@ -574,7 +574,15 @@ namespace RandomX {
 
 	void JitCompilerX86::h_FPROUND(Instruction& instr, int i) {
 		genar(instr);
-		emit(0x00250de0c1c88b48); //mov rcx,rax; shl eax,0xd
+		emitByte(0x48);
+		emit(uint16_t(0xc88b)); //mov rcx,rax
+		int rotate = (13 - (instr.imm8 & 63)) & 63;
+		if (rotate != 0) {
+			emitByte(0x48);
+			emit(uint16_t(0xc0c1)); //rol rax
+			emitByte(rotate);
+		}
+		emit(uint16_t(0x0025));
 		emit(0x00009fc00d000060); //and eax,0x6000; or eax,0x9fc0
 		emit(0x2454ae0ff8244489); //ldmxcsr DWORD PTR [rsp-0x8]
 		emitByte(0xf8);

src/program.inc

@@ -8859,7 +8859,7 @@ rx_body_509:
 	and ecx, 2047
 	mov rax, qword ptr [rsi+rcx*8]
 	mov rcx, rax
-	shl eax, 13
+	rol rax, 34
 	and eax, 24576
 	or eax, 40896
 	mov dword ptr [rsp - 8], eax