Some cleanup related to toolchain, add Zbr

BitManipZbr.scala

@@ -0,0 +1,134 @@
+// WARNING: this is auto-generated code!
+// See https://github.com/rdolbeau/VexRiscvBPluginGenerator/
+package vexriscv.plugin
+import spinal.core._
+import vexriscv.{Stageable, DecoderService, VexRiscv}
+object BitManipZbrPlugin {
+	object BitManipZbrCtrlCRC32xEnum extends SpinalEnum(binarySequential) {
+		 val CTRL_CRC32xdotB, CTRL_CRC32xdotH, CTRL_CRC32xdotW = newElement()
+	}
+	object BitManipZbrCtrlEnum extends SpinalEnum(binarySequential) {
+		 val CTRL_CRC32x = newElement()
+	}
+	object BitManipZbrCtrlCRC32x extends Stageable(BitManipZbrCtrlCRC32xEnum())
+	object BitManipZbrCtrl extends Stageable(BitManipZbrCtrlEnum())
+// Prologue
+
+	def fun_crc32xdotb(rs1: Bits, isC : Bool) : Bits = {
+	   val p = ((isC === True) ? (B"32'x82F63B78") | (B"32'xEDB88320"))
+	   var x = rs1
+	   for (i <- 0 to 7) {
+	   	   x = (x |>> 1) ^ ((x(0) === True) ? (p) | (B"32'x00000000"))
+	   }
+	   val r = x
+	   r // return value
+	}
+	def fun_crc32xdoth(rs1: Bits, isC : Bool) : Bits = {
+	   val p = ((isC === True) ? (B"32'x82F63B78") | (B"32'xEDB88320"))
+	   var x = rs1
+	   for (i <- 0 to 15) {
+	   	   x = (x |>> 1) ^ ((x(0) === True) ? (p) | (B"32'x00000000"))
+	   }
+	   val r = x
+	   r // return value
+	}
+	def fun_crc32xdotw(rs1: Bits, isC : Bool) : Bits = {
+	   val p = ((isC === True) ? (B"32'x82F63B78") | (B"32'xEDB88320"))
+	   var x = rs1
+	   for (i <- 0 to 31) {
+	   	   x = (x |>> 1) ^ ((x(0) === True) ? (p) | (B"32'x00000000"))
+	   }
+	   val r = x
+	   r // return value
+	}
+
+// End prologue
+} // object Plugin
+class BitManipZbrPlugin(earlyInjection : Boolean = true) extends Plugin[VexRiscv] {
+	import BitManipZbrPlugin._
+	object IS_BitManipZbr extends Stageable(Bool)
+	object BitManipZbr_FINAL_OUTPUT extends Stageable(Bits(32 bits))
+	override def setup(pipeline: VexRiscv): Unit = {
+		import pipeline.config._
+		val immediateActions = List[(Stageable[_ <: BaseType],Any)](
+			SRC1_CTRL                -> Src1CtrlEnum.RS,
+			SRC2_CTRL                -> Src2CtrlEnum.IMI,
+			REGFILE_WRITE_VALID      -> True,
+			BYPASSABLE_EXECUTE_STAGE -> Bool(earlyInjection),
+			BYPASSABLE_MEMORY_STAGE  -> True,
+			RS1_USE -> True,
+			IS_BitManipZbr -> True
+			)
+		val binaryActions = List[(Stageable[_ <: BaseType],Any)](
+			SRC1_CTRL                -> Src1CtrlEnum.RS,
+			SRC2_CTRL                -> Src2CtrlEnum.RS,
+			REGFILE_WRITE_VALID      -> True,
+			BYPASSABLE_EXECUTE_STAGE -> Bool(earlyInjection),
+			BYPASSABLE_MEMORY_STAGE  -> True,
+			RS1_USE -> True,
+			RS2_USE -> True,
+			IS_BitManipZbr -> True
+			)
+		val unaryActions = List[(Stageable[_ <: BaseType],Any)](
+			SRC1_CTRL                -> Src1CtrlEnum.RS,
+			REGFILE_WRITE_VALID      -> True,
+			BYPASSABLE_EXECUTE_STAGE -> Bool(earlyInjection),
+			BYPASSABLE_MEMORY_STAGE  -> True,
+			RS1_USE -> True,
+			IS_BitManipZbr -> True
+			)
+		val ternaryActions = List[(Stageable[_ <: BaseType],Any)](
+			SRC1_CTRL                -> Src1CtrlEnum.RS,
+			SRC2_CTRL                -> Src2CtrlEnum.RS,
+			SRC3_CTRL                -> Src3CtrlEnum.RS,
+			REGFILE_WRITE_VALID      -> True,
+			BYPASSABLE_EXECUTE_STAGE -> Bool(earlyInjection),
+			BYPASSABLE_MEMORY_STAGE  -> True,
+			RS1_USE -> True,
+			RS2_USE -> True,
+			RS3_USE -> True,
+			IS_BitManipZbr -> True
+			)
+		val immTernaryActions = List[(Stageable[_ <: BaseType],Any)](
+			SRC1_CTRL                -> Src1CtrlEnum.RS,
+			SRC2_CTRL                -> Src2CtrlEnum.IMI,
+			SRC3_CTRL                -> Src3CtrlEnum.RS,
+			REGFILE_WRITE_VALID      -> True,
+			BYPASSABLE_EXECUTE_STAGE -> Bool(earlyInjection),
+			BYPASSABLE_MEMORY_STAGE  -> True,
+			RS1_USE -> True,
+			RS3_USE -> True,
+			IS_BitManipZbr -> True
+			)
+		def CRC32xdotB_KEY = M"01100001-000-----001-----0010011"
+		def CRC32xdotH_KEY = M"01100001-001-----001-----0010011"
+		def CRC32xdotW_KEY = M"01100001-010-----001-----0010011"
+		val decoderService = pipeline.service(classOf[DecoderService])
+		decoderService.addDefault(IS_BitManipZbr, False)
+		decoderService.add(List(
+			CRC32xdotB_KEY	-> (unaryActions ++ List(BitManipZbrCtrl -> BitManipZbrCtrlEnum.CTRL_CRC32x, BitManipZbrCtrlCRC32x -> BitManipZbrCtrlCRC32xEnum.CTRL_CRC32xdotB)),
+			CRC32xdotH_KEY	-> (unaryActions ++ List(BitManipZbrCtrl -> BitManipZbrCtrlEnum.CTRL_CRC32x, BitManipZbrCtrlCRC32x -> BitManipZbrCtrlCRC32xEnum.CTRL_CRC32xdotH)),
+			CRC32xdotW_KEY	-> (unaryActions ++ List(BitManipZbrCtrl -> BitManipZbrCtrlEnum.CTRL_CRC32x, BitManipZbrCtrlCRC32x -> BitManipZbrCtrlCRC32xEnum.CTRL_CRC32xdotW))
+		))
+	} // override def setup
+	override def build(pipeline: VexRiscv): Unit = {
+		import pipeline._
+		import pipeline.config._
+		execute plug new Area{
+			import execute._
+			val val_CRC32x = input(BitManipZbrCtrlCRC32x).mux(
+				BitManipZbrCtrlCRC32xEnum.CTRL_CRC32xdotB -> fun_crc32xdotb(input(SRC1), input(INSTRUCTION)(23)).asBits,
+				BitManipZbrCtrlCRC32xEnum.CTRL_CRC32xdotH -> fun_crc32xdoth(input(SRC1), input(INSTRUCTION)(23)).asBits,
+				BitManipZbrCtrlCRC32xEnum.CTRL_CRC32xdotW -> fun_crc32xdotw(input(SRC1), input(INSTRUCTION)(23)).asBits
+			) // mux CRC32x
+			insert(BitManipZbr_FINAL_OUTPUT) := val_CRC32x.asBits
+		} // execute plug newArea
+		val injectionStage = if(earlyInjection) execute else memory
+		injectionStage plug new Area {
+			import injectionStage._
+			when (arbitration.isValid && input(IS_BitManipZbr)) {
+				output(REGFILE_WRITE_DATA) := input(BitManipZbr_FINAL_OUTPUT)
+			} // when input is
+		} // injectionStage plug newArea
+	} // override def build
+} // class Plugin

Makefile

@@ -80,7 +80,8 @@ BitManipZbbZbp.scala: gen_plugin data_bitmanip.txt
 BitManipZbp.scala: gen_plugin data_bitmanip.txt
 	./gen_plugin -n BitManipZbp -i data_bitmanip.txt -I Zbp >| $@
 
-## Zbr unimplemented, crc32* are missing
+BitManipZbr.scala: gen_plugin data_crc.txt
+	./gen_plugin -n BitManipZbr -i data_crc.txt -I Zbr >| $@
 
 BitManipZbs.scala: gen_plugin data_bitmanip.txt
 	./gen_plugin -n BitManipZbs -i data_bitmanip.txt -I Zbs >| $@
@@ -127,7 +128,7 @@ PackedSIMDSlow.scala: gen_plugin data_Zpn_2cycles.txt
 PackedSIMDWide.scala: gen_plugin data_Zp64.txt
 	./gen_plugin -w -n PackedSIMDWide -i data_Zp64.txt -I '*' >| $@
 
-B: BitManipZba.scala BitManipZbb.scala BitManipZbbZbp.scala BitManipZbc.scala BitManipZbe1cycle.scala BitManipZbe2cycles.scala BitManipZbf.scala BitManipBFPonly.scala BitManipZbp.scala BitManipZbs.scala BitManipZbt.scala
+B: BitManipZba.scala BitManipZbb.scala BitManipZbbZbp.scala BitManipZbc.scala BitManipZbe1cycle.scala BitManipZbe2cycles.scala BitManipZbf.scala BitManipBFPonly.scala BitManipZbp.scala BitManipZbr.scala BitManipZbs.scala BitManipZbt.scala
 
 P: PackedSIMDBase.scala PackedSIMDSlow.scala PackedSIMDWide.scala
 

README.md

@@ -10,7 +10,6 @@ The generated plugin is for RV32 only. It doesn't yet support all B instructions
 
 * all instructions ending in 'W', as they are RV64-only
 * BMAT*, as they are RV64-only
-* CRC32*
 * Three-operands instructions (CMIX, CMOV, FS[RL]*); they are available but need a VexRiscv patch to support the third input (all VexRiscv patch are available on https://github.com/rdolbeau/VexRiscv/tree/three_operands)
 
 There is support for partial instructions (rev8, zext.h, orc.b) so that the default plugins generated for Zba, Zbb and Zbc should be feature-complete. To get everything without conflicts, use:

data_crc.txt

@@ -0,0 +1,40 @@
+// Zbr
+I	CRC32xdotB	 CRC32xdotB	01100001-000-----001-----0010011 CRC32x	Zbr
+I	CRC32xdotH	 CRC32xdotH	01100001-001-----001-----0010011 CRC32x	Zbr
+I	CRC32xdotW	 CRC32xdotW	01100001-010-----001-----0010011 CRC32x	Zbr
+//I	CRC32xdotD	 CRC32xdotD	01100001-011-----001-----0010011 CRC32x	Zbr
+
+S	CRC32xdotB  "fun_crc32xdotb(input(SRC1), input(INSTRUCTION)(23))"
+S	CRC32xdotH  "fun_crc32xdoth(input(SRC1), input(INSTRUCTION)(23))"
+S	CRC32xdotW  "fun_crc32xdotw(input(SRC1), input(INSTRUCTION)(23))"
+
+P	"""
+	def fun_crc32xdotb(rs1: Bits, isC : Bool) : Bits = {
+	   val p = ((isC === True) ? (B"32'x82F63B78") | (B"32'xEDB88320"))
+	   var x = rs1
+	   for (i <- 0 to 7) {
+	   	   x = (x |>> 1) ^ ((x(0) === True) ? (p) | (B"32'x00000000"))
+	   }
+	   val r = x
+	   r // return value
+	}
+	def fun_crc32xdoth(rs1: Bits, isC : Bool) : Bits = {
+	   val p = ((isC === True) ? (B"32'x82F63B78") | (B"32'xEDB88320"))
+	   var x = rs1
+	   for (i <- 0 to 15) {
+	   	   x = (x |>> 1) ^ ((x(0) === True) ? (p) | (B"32'x00000000"))
+	   }
+	   val r = x
+	   r // return value
+	}
+	def fun_crc32xdotw(rs1: Bits, isC : Bool) : Bits = {
+	   val p = ((isC === True) ? (B"32'x82F63B78") | (B"32'xEDB88320"))
+	   var x = rs1
+	   for (i <- 0 to 31) {
+	   	   x = (x |>> 1) ^ ((x(0) === True) ? (p) | (B"32'x00000000"))
+	   }
+	   val r = x
+	   r // return value
+	}
+"""
+

new_instructions_support_b.h

@@ -5,21 +5,6 @@
 
 #include "new_instructions_support.h" // for support macros
 
-// for instructions currently missing in the toolchain
-ASM2MACRO(XPERM_N,0x28002033)
-ASM2MACRO(XPERM_B,0x28004033)
-ASM2MACRO(XPERM_H,0x28006033)
-ASM2MACRO(_SH1ADD,0x20002033)
-ASM2MACRO(_SH2ADD,0x20004033)
-ASM2MACRO(_SH3ADD,0x20006033)
-FUN2(xperm_n,XPERM_N)
-FUN2(xperm_b,XPERM_B)
-FUN2(xperm_h,XPERM_H)
-FUN2(_sh1add,_SH1ADD)
-FUN2(_sh2add,_SH2ADD)
-FUN2(_sh3add,_SH3ADD)
-
-
 #ifdef ENABLE_XAR
 ASM3RMACRO(XAR, 0xc8000077)
 FUN3R(__rv__xar, XAR)

r5.mk

@@ -1,5 +1,5 @@
 ## B Toolchain so we get some automatic B generation
-R5B_TOOLCHAIN=/opt/riscv64b/
+R5B_TOOLCHAIN=/opt/riscv64bk/
 R5B_GCC=$(R5B_TOOLCHAIN)/bin/riscv64-unknown-elf-gcc
 R5B_OPT=-Os -march=rv32imab -mabi=ilp32 -I.
 

test_b.c

@@ -36,38 +36,40 @@
 #include "new_instructions_support_b.h"
 
 #define _rv64_clmul2(a,b) _rv64_clmul(a,b)
+
+
+			//not yet in rvintrin.h ?
+static inline uint_xlen_t _rv32_sh1add(uint_xlen_t rs1, uint_xlen_t rs2) {
+	uint_xlen_t rd;
+	asm ("sh1add %0, %1, %2\n" : "=r" (rd) : "r" (rs1), "r" (rs2));
+	return rd;
+}
+static inline uint_xlen_t _rv32_sh2add(uint_xlen_t rs1, uint_xlen_t rs2) {
+	uint_xlen_t rd;
+	asm ("sh2add %0, %1, %2\n" : "=r" (rd) : "r" (rs1), "r" (rs2));
+	return rd;
+}
+static inline uint_xlen_t _rv32_sh3add(uint_xlen_t rs1, uint_xlen_t rs2) {
+	uint_xlen_t rd;
+	asm ("sh3add %0, %1, %2\n" : "=r" (rd) : "r" (rs1), "r" (rs2));
+	return rd;
+}
 #else
 #include <rvintrin.h> // emulation
 
 typedef uint32_t uint_xlen_t;
 #define XLEN 32
-uint_xlen_t xperm(uint_xlen_t rs1, uint_xlen_t rs2, int sz_log2)
-{
-    uint_xlen_t r = 0;
-    uint_xlen_t sz = 1LL << sz_log2;
-    uint_xlen_t mask = (1LL << sz) - 1;
-    for (int i = 0; i < XLEN; i += sz) {
-        uint_xlen_t pos = ((rs2 >> i) & mask) << sz_log2;
-        if (pos < XLEN)
-            r |= ((rs1 >> pos) & mask) << i;
-    }
-    return r;
-}
-uint_xlen_t xperm_n (uint_xlen_t rs1, uint_xlen_t rs2) {  return xperm(rs1, rs2, 2); }
-uint_xlen_t xperm_b (uint_xlen_t rs1, uint_xlen_t rs2) {  return xperm(rs1, rs2, 3); }
-uint_xlen_t xperm_h (uint_xlen_t rs1, uint_xlen_t rs2) {  return xperm(rs1, rs2, 4); }
-uint_xlen_t xperm_w (uint_xlen_t rs1, uint_xlen_t rs2) {  return xperm(rs1, rs2, 5); }
-
   
-uint_xlen_t _sh1add(uint_xlen_t rs1, uint_xlen_t rs2)
+			//not yet in rvintrin.h ?
+uint_xlen_t _rv32_sh1add(uint_xlen_t rs1, uint_xlen_t rs2)
 {
     return (rs1 << 1) + rs2;
 }
-uint_xlen_t _sh2add(uint_xlen_t rs1, uint_xlen_t rs2)
+uint_xlen_t _rv32_sh2add(uint_xlen_t rs1, uint_xlen_t rs2)
 {
     return (rs1 << 2) + rs2;
 }
-uint_xlen_t _sh3add(uint_xlen_t rs1, uint_xlen_t rs2)
+uint_xlen_t _rv32_sh3add(uint_xlen_t rs1, uint_xlen_t rs2)
 {
     return (rs1 << 3) + rs2;
 }
@@ -118,7 +120,14 @@ int main(int argc, char **argv) {
 		T1(_rv32_clz);
 		T1(_rv32_ctz);
 		T1(_rv32_pcnt);
-			
+
+		T1(_rv_crc32_b);
+		T1(_rv_crc32_h);
+		T1(_rv_crc32_w);
+		T1(_rv_crc32c_b);
+		T1(_rv_crc32c_h);
+		T1(_rv_crc32c_w);
+
 		for (index1 = 0 ; index1 < nonrandom_cnt[1] ; index1++) {
 			b = nonrandom_b[index1];
   
@@ -139,10 +148,6 @@ int main(int argc, char **argv) {
 			T2(_rv_xnor);
 			T2(_rv_orn);
 
-			//T2(_rv32_sh1add);
-			//T2(_rv32_sh2add);
-			//T2(_rv32_sh3add);
-
 			T2(_rv32_sbset);
 			T2(_rv32_sbclr);
 			T2(_rv32_sbinv);
@@ -152,9 +157,10 @@ int main(int argc, char **argv) {
 			T2(_rv32_minu);
 			T2(_rv32_max);
 			T2(_rv32_maxu);
-  
-			T2(_rv32_slo);
-			T2(_rv32_sro);
+
+			// no longer in toolchain, dropped from extension
+			/* T2(_rv32_slo); */
+			/* T2(_rv32_sro); */
 
 			//T2(_rv32_xperm_b);
 
@@ -175,14 +181,13 @@ int main(int argc, char **argv) {
 					printf("0x%016llx 0x%016llx (0x%016llx)\n", z, z2, z^z2);
 				}
 
-			// extra stuff
-			T2(_sh1add);
-			T2(_sh2add);
-			T2(_sh3add);
+			T2(_rv32_sh1add);
+			T2(_rv32_sh2add);
+			T2(_rv32_sh3add);
 
-			T2(xperm_n);
-			T2(xperm_b);
-			T2(xperm_h);
+			T2(_rv_xperm_n);
+			T2(_rv_xperm_b);
+			T2(_rv_xperm_h);
 
 			T2(_rv32_bfp);