Fix FPU x0 writeback scoreboard

hw/rtl/fpu/VX_fpu_unit.sv

@@ -259,7 +259,9 @@ module VX_fpu_unit import VX_gpu_pkg::*, VX_fpu_pkg::*; #(
             .valid_out (per_block_result_if[block_idx].valid),
             .ready_out (per_block_result_if[block_idx].ready)
         );
-        assign per_block_result_if[block_idx].data.wb = 1'b1;
+        // Avoid writeback to x0, which scoreboard does not reserve.
+        assign per_block_result_if[block_idx].data.wb =
+            fpu_rsp_rd != make_reg_num(REG_TYPE_I, RV_REGS_BITS'(0));
     end
 
     VX_gather_unit #(

tests/kernel/conform/main.cpp

@@ -28,11 +28,13 @@ int main() {
 
 	errors += test_shfl();
 
+	errors += test_feq_x0_scoreboard();
+
 	if (0 == errors) {
 		PRINTF("Passed!\n");
 	} else {
 		PRINTF("Failed!\n");
 	}
 
 	return errors;
-}
+}

tests/kernel/conform/tests.cpp

@@ -31,6 +31,28 @@ int __attribute__((noinline)) make_full_tmask(int num_threads) {
 
 ///////////////////////////////////////////////////////////////////////////////
 
+void __attribute__((noinline)) do_feq_x0_scoreboard() {
+	asm volatile(
+		"fsgnj.s f24, f0, f0\n"
+		"fsgnj.s f1, f0, f0\n"
+		"feq.s x0, f24, f1\n"
+		:
+		:
+		: "memory");
+}
+
+int test_feq_x0_scoreboard() {
+	PRINTF("FEQ x0 Scoreboard Test\n");
+	int num_threads = std::min(vx_num_threads(), 4);
+	int tmask = make_full_tmask(num_threads);
+	vx_tmc(tmask);
+	do_feq_x0_scoreboard();
+	vx_tmc_one();
+	return 0;
+}
+
+///////////////////////////////////////////////////////////////////////////////
+
 #define GLOBAL_MEM_SZ 8
 int global_buffer[GLOBAL_MEM_SZ];
 
@@ -401,4 +423,4 @@ int test_shfl() {
 	do_shfl();
   vx_tmc_one(); // back to thread0
 	return check_error(shfl_buffer, 0, num_threads);
-}
+}

tests/kernel/conform/tests.h

@@ -3,6 +3,8 @@
 
 #define PRINTF vx_printf
 
+int test_feq_x0_scoreboard();
+
 int test_global_memory();
 
 int test_local_memory();