From 1f99de511760b3acb546e62623c65d6143b30a77 Mon Sep 17 00:00:00 2001
From: Clifford Wolf <clifford@clifford.at>
Date: Sun, 28 Jun 2015 13:07:50 +0200
Subject: [PATCH] Improvements in picorv32_pcpi_mul

---
 README.md        |  3 +++
 firmware/start.S |  4 +++
 picorv32.v       | 63 +++++++++++++++++++++++++++++++++---------------
 testbench.v      |  4 ++-
 4 files changed, 53 insertions(+), 21 deletions(-)

diff --git a/README.md b/README.md
index 3b84a19..cc6bf91 100644
--- a/README.md
+++ b/README.md
@@ -194,6 +194,9 @@ CPI numbers for a core built without ENABLE_REGS_DUALPORT.
 | indirect jump (jalr) |    6 |        6 |
 | shift operations     | 4-14 |     4-15 |
 
+When `ENABLE_MUL` is activated, then a `MUL` instruction will execute
+in 42 cycles and a `MULH[SU|U]` instruction will execute in 74 cycles.
+
 Dhrystone benchmark results: 0.309 DMIPS/MHz (544 Dhrystones/Second/MHz)
 
 For the Dhrystone benchmark the average CPI is 4.167.
diff --git a/firmware/start.S b/firmware/start.S
index f39665e..278f0d4 100644
--- a/firmware/start.S
+++ b/firmware/start.S
@@ -292,6 +292,10 @@ start:
 	/* break */
 	sbreak
 
+
+/* Hard mul functions for multest.c
+ **********************************/
+
 hard_mul:
 	mul a0, a0, a1
 	ret
diff --git a/picorv32.v b/picorv32.v
index a311080..abccc23 100644
--- a/picorv32.v
+++ b/picorv32.v
@@ -660,12 +660,6 @@ module picorv32 #(
 				reg_pc <= current_pc;
 				reg_next_pc <= current_pc;
 
-				if (WITH_PCPI) begin
-					pcpi_insn_valid <= 0;
-					pcpi_rs1_valid <= 0;
-					pcpi_rs2_valid <= 0;
-				end
-
 				latched_store <= 0;
 				latched_stalu <= 0;
 				latched_branch <= 0;
@@ -726,6 +720,10 @@ module picorv32 #(
 							reg_sh <= decoded_rs2 ? cpuregs[decoded_rs2] : 0;
 							reg_op2 <= decoded_rs2 ? cpuregs[decoded_rs2] : 0;
 							if (pcpi_int_ready) begin
+								mem_do_rinst <= 1;
+								pcpi_insn_valid <= 0;
+								pcpi_rs1_valid <= 0;
+								pcpi_rs2_valid <= 0;
 								reg_out <= pcpi_int_rd;
 								latched_store <= pcpi_int_rd_valid;
 								cpu_state <= cpu_state_fetch;
@@ -848,6 +846,10 @@ module picorv32 #(
 				if (WITH_PCPI && pcpi_insn_valid) begin
 					pcpi_rs2_valid <= 1;
 					if (pcpi_int_ready) begin
+						mem_do_rinst <= 1;
+						pcpi_insn_valid <= 0;
+						pcpi_rs1_valid <= 0;
+						pcpi_rs2_valid <= 0;
 						reg_out <= pcpi_int_rd;
 						latched_store <= pcpi_int_rd_valid;
 						cpu_state <= cpu_state_fetch;
@@ -1023,7 +1025,10 @@ endmodule
  * picorv32_pcpi_mul
  ***************************************************************/
 
-module picorv32_pcpi_mul (
+module picorv32_pcpi_mul #(
+	// increasing this parameter increases performance and core size
+	parameter STEPS_AT_ONCE = 1
+) (
 	input clk, resetn,
 
 	input             pcpi_insn_valid,
@@ -1067,9 +1072,32 @@ module picorv32_pcpi_mul (
 	end
 
 	reg [63:0] rs1, rs2, rd, rdx;
+	reg [63:0] next_rs1, next_rs2, next_rd, next_rdx, next_rdt;
 	reg [6:0] mul_counter;
 	reg mul_waiting;
 	reg mul_finish;
+	integer i;
+
+	// carry save accumulator
+	always @* begin
+		next_rd = rd;
+		next_rdx = rdx;
+		next_rs1 = rs1;
+		next_rs2 = rs2;
+
+		for (i = 0; i < STEPS_AT_ONCE; i=i+1) begin
+			if (next_rs1[0]) begin
+				next_rdt = (next_rd ^ next_rdx) ^ next_rs2;
+				next_rdx = ((next_rd & next_rdx) | (next_rd & next_rs2) | (next_rdx & next_rs2)) << 1;
+			end else begin
+				next_rdt = next_rd ^ next_rdx;
+				next_rdx = (next_rd & next_rdx) << 1;
+			end
+			next_rd = next_rdt;
+			next_rs1 = next_rs1 >> 1;
+			next_rs2 = next_rs2 << 1;
+		end
+	end
 
 	always @(posedge clk) begin
 		mul_finish <= 0;
@@ -1089,21 +1117,16 @@ module picorv32_pcpi_mul (
 
 			rd <= 0;
 			rdx <= 0;
-			mul_counter <= instr_any_mulh ? 64 : 32;
+			mul_counter <= (instr_any_mulh ? 63 - STEPS_AT_ONCE : 31 - STEPS_AT_ONCE);
 			mul_waiting <= !mul_start;
 		end else begin
-			// carry save accumulator
-			if (rs1[0]) begin
-				rd <= rd ^ rdx ^ rs2;
-				rdx <= ((rd & rdx) | (rd & rs2) | (rdx & rs2)) << 1;
-			end else begin
-				rd <= rd ^ rdx;
-				rdx <= (rd & rdx) << 1;
-			end
-			rs1 <= rs1 >> 1;
-			rs2 <= rs2 << 1;
-			mul_counter <= mul_counter - 1;
-			if (!mul_counter) begin
+			rd <= next_rd;
+			rdx <= next_rdx;
+			rs1 <= next_rs1;
+			rs2 <= next_rs2;
+
+			mul_counter <= mul_counter - STEPS_AT_ONCE;
+			if (mul_counter[6]) begin
 				mul_finish <= 1;
 				mul_waiting <= 1;
 			end
diff --git a/testbench.v b/testbench.v
index 135fbdc..53badd3 100644
--- a/testbench.v
+++ b/testbench.v
@@ -244,10 +244,12 @@ module testbench;
 		$finish;
 	end
 
+	integer cycle_counter;
 	always @(posedge clk) begin
+		cycle_counter <= resetn ? cycle_counter + 1 : 0;
 		if (resetn && trap) begin
 			repeat (10) @(posedge clk);
-			$display("TRAP");
+			$display("TRAP after %1d clock cycles", cycle_counter);
 			$finish;
 		end
 	end