xiaoxiang781216 commented on a change in pull request #5729:
URL: https://github.com/apache/incubator-nuttx/pull/5729#discussion_r825389413
##########
File path: arch/xtensa/src/common/xtensa_asm_utils.h
##########
@@ -0,0 +1,76 @@
+/****************************************************************************
+ * arch/xtensa/src/common/xtensa_asm_utils.h
+ *
+ * Copyright (c) 2017, Intel Corporation
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ *
+ ****************************************************************************/
+
+#ifndef __ARCH_XTENSA_SRC_COMMON_XTENSA_ASM_UTILS_H
+#define __ARCH_XTENSA_SRC_COMMON_XTENSA_ASM_UTILS_H
+
+/****************************************************************************
+ * Assembly Language Macros
+ ****************************************************************************/
+
+/****************************************************************************
+ *
+ * Name: SPILL_ALL_WINDOWS
+ *
+ * Spills all windowed registers (i.e. registers not visible as
+ * A0-A15) to their ABI-defined spill regions on the stack.
+ *
+ * Unlike the Xtensa HAL implementation, this code requires that the
+ * EXCM and WOE bit be enabled in PS, and relies on repeated hardware
+ * exception handling to do the register spills. The trick is to do a
+ * noop write to the high registers, which the hardware will trap
+ * (into an overflow exception) in the case where those registers are
+ * already used by an existing call frame. Then it rotates the window
+ * and repeats until all but the A0-A3 registers of the original frame
+ * are guaranteed to be spilled, eventually rotating back around into
+ * the original frame. Advantages:
+ *
+ * - Vastly smaller code size
+ *
+ * - More easily maintained if changes are needed to window over/underflow
+ * exception handling.
+ *
+ * - Requires no scratch registers to do its work, so can be used safely in
+ * any context.
+ *
+ * - If the WOE bit is not enabled (for example, in code written for
+ * the CALL0 ABI), this becomes a silent noop and operates compatbily.
+ *
+ * - Hilariously it's ACTUALLY FASTER than the HAL routine. And not
+ * just a little bit, it's MUCH faster. With a mostly full register
+ * file on an LX6 core (ESP-32) I'm measuring 145 cycles to spill
+ * registers with this vs. 279 (!) to do it with
+ * xthal_spill_windows().
+ ****************************************************************************/
+
+.macro SPILL_ALL_WINDOWS
Review comment:
why not put into xtensa_context.S? If only xtensa_context.S will call
SPILL_ALL_WINDOWS?
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscr...@nuttx.apache.org
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
