linux-riscv.lists.infradead.org archive mirror
 help / color / mirror / Atom feed
From: Charlie Jenkins <charlie@rivosinc.com>
To: Charlie Jenkins <charlie@rivosinc.com>,
	 Palmer Dabbelt <palmer@dabbelt.com>,
	Conor Dooley <conor@kernel.org>,
	 Samuel Holland <samuel.holland@sifive.com>,
	linux-riscv@lists.infradead.org,  linux-kernel@vger.kernel.org
Cc: Paul Walmsley <paul.walmsley@sifive.com>,
	 Albert Ou <aou@eecs.berkeley.edu>
Subject: [PATCH v2 3/5] riscv: Vector checksum header
Date: Tue, 05 Sep 2023 21:46:52 -0700	[thread overview]
Message-ID: <20230905-optimize_checksum-v2-3-ccd658db743b@rivosinc.com> (raw)
In-Reply-To: <20230905-optimize_checksum-v2-0-ccd658db743b@rivosinc.com>

Vector code is written in assembly rather than using the GCC vector
instrinsics because they did not provide optimal code. Vector
instrinsic types are still used so the inline assembly can
appropriately select vector registers. However, this code cannot be
merged yet because it is currently not possible to use vector
instrinsics in the kernel because vector support needs to be directly
enabled by assembly.

Signed-off-by: Charlie Jenkins <charlie@rivosinc.com>
---
 arch/riscv/include/asm/checksum.h | 87 +++++++++++++++++++++++++++++++++++++++
 1 file changed, 87 insertions(+)

diff --git a/arch/riscv/include/asm/checksum.h b/arch/riscv/include/asm/checksum.h
index 3f9d5a202e95..1d6c23cd1221 100644
--- a/arch/riscv/include/asm/checksum.h
+++ b/arch/riscv/include/asm/checksum.h
@@ -10,6 +10,10 @@
 #include <linux/in6.h>
 #include <linux/uaccess.h>
 
+#ifdef CONFIG_RISCV_ISA_V
+#include <riscv_vector.h>
+#endif
+
 #ifdef CONFIG_32BIT
 typedef unsigned int csum_t;
 #else
@@ -43,6 +47,89 @@ static inline __sum16 csum_fold(__wsum sum)
  */
 static inline __sum16 ip_fast_csum(const void *iph, unsigned int ihl)
 {
+#ifdef CONFIG_RISCV_ISA_V
+	if (IS_ENABLED(CONFIG_RISCV_ALTERNATIVE)) {
+		/*
+		 * Vector is likely available when the kernel is compiled with
+		 * vector support, so nop when vector is available and jump when
+		 * vector is not available.
+		 */
+		asm_volatile_goto(ALTERNATIVE("j %l[no_vector]", "nop", 0,
+					      RISCV_ISA_EXT_v, 1)
+				  :
+				  :
+				  :
+				  : no_vector);
+	} else {
+		if (!__riscv_isa_extension_available(NULL, RISCV_ISA_EXT_v))
+			goto no_vector;
+	}
+
+	vuint64m1_t prev_buffer;
+	vuint32m1_t curr_buffer;
+	unsigned int vl;
+#ifdef CONFIG_32_BIT
+	csum_t high_result, low_result;
+
+	riscv_v_enable();
+	asm(".option push						\n\
+	.option arch, +v						\n\
+	vsetivli x0, 1, e64, ta, ma					\n\
+	vmv.v.i %[prev_buffer], 0					\n\
+	1:								\n\
+	vsetvli %[vl], %[ihl], e32, m1, ta, ma				\n\
+	vle32.v %[curr_buffer], (%[iph])				\n\
+	vwredsumu.vs %[prev_buffer], %[curr_buffer], %[prev_buffer]	\n\
+	sub %[ihl], %[ihl], %[vl]					\n\
+	slli %[vl], %[vl], 2						\n\
+	add %[iph], %[vl], %[iph]					\n\
+	# If not all of iph could fit into vector reg, do another sum	\n\
+	bne %[ihl], zero, 1b						\n\
+	vsetivli x0, 1, e64, m1, ta, ma					\n\
+	vmv.x.s %[low_result], %[prev_buffer]				\n\
+	addi %[vl], x0, 32						\n\
+	vsrl.vx %[prev_buffer], %[prev_buffer], %[vl]			\n\
+	vmv.x.s %[high_result], %[prev_buffer]				\n\
+	.option pop"
+	: [vl] "=&r" (vl), [prev_buffer] "=&vd" (prev_buffer),
+		[curr_buffer] "=&vd" (curr_buffer),
+		[high_result] "=&r" (high_result),
+		[low_result] "=&r" (low_result)
+	: [iph] "r" (iph), [ihl] "r" (ihl));
+	riscv_v_disable();
+
+	high_result += low_result;
+	high_result += high_result < low_result;
+#else // !CONFIG_32_BIT
+	csum_t result;
+
+	riscv_v_enable();
+	asm(".option push						\n\
+	.option arch, +v						\n\
+	vsetivli x0, 1, e64, ta, ma					\n\
+	vmv.v.i %[prev_buffer], 0					\n\
+	1:								\n\
+	# Setup 32-bit sum of iph					\n\
+	vsetvli %[vl], %[ihl], e32, m1, ta, ma				\n\
+	vle32.v %[curr_buffer], (%[iph])				\n\
+	# Sum each 32-bit segment of iph that can fit into a vector reg	\n\
+	vwredsumu.vs %[prev_buffer], %[curr_buffer], %[prev_buffer]     \n\
+	subw %[ihl], %[ihl], %[vl]					\n\
+	slli %[vl], %[vl], 2						\n\
+	addw %[iph], %[vl], %[iph]					\n\
+	# If not all of iph could fit into vector reg, do another sum	\n\
+	bne %[ihl], zero, 1b						\n\
+	vsetvli x0, x0, e64, m1, ta, ma					\n\
+	vmv.x.s %[result], %[prev_buffer]				\n\
+	.option pop"
+	: [vl] "=&r" (vl), [prev_buffer] "=&vd" (prev_buffer),
+		[curr_buffer] "=&vd" (curr_buffer), [result] "=&r" (result)
+	: [iph] "r" (iph), [ihl] "r" (ihl));
+	riscv_v_disable();
+#endif // !CONFIG_32_BIT
+no_vector:
+#endif // !CONFIG_RISCV_ISA_V
+
 	csum_t csum = 0;
 	int pos = 0;
 

-- 
2.42.0


_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv

  parent reply	other threads:[~2023-09-06  4:47 UTC|newest]

Thread overview: 16+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2023-09-06  4:46 [PATCH v2 0/5] riscv: Add fine-tuned checksum functions Charlie Jenkins
2023-09-06  4:46 ` [PATCH v2 1/5] riscv: Checksum header Charlie Jenkins
2023-09-07  9:40   ` Conor Dooley
2023-09-07 17:44     ` Charlie Jenkins
2023-09-10 21:20       ` David Laight
2023-09-11 18:16         ` Charlie Jenkins
2023-09-06  4:46 ` [PATCH v2 2/5] riscv: Add checksum library Charlie Jenkins
2023-09-07  9:52   ` Conor Dooley
2023-09-07 17:47     ` Charlie Jenkins
2023-09-06  4:46 ` Charlie Jenkins [this message]
2023-09-07  9:47   ` [PATCH v2 3/5] riscv: Vector checksum header Conor Dooley
2023-09-07 17:43     ` Charlie Jenkins
2023-09-07  9:58   ` Conor Dooley
2023-09-07 17:41     ` Charlie Jenkins
2023-09-06  4:46 ` [PATCH v2 4/5] riscv: Vector checksum library Charlie Jenkins
2023-09-06  4:46 ` [PATCH v2 5/5] riscv: Test checksum functions Charlie Jenkins

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20230905-optimize_checksum-v2-3-ccd658db743b@rivosinc.com \
    --to=charlie@rivosinc.com \
    --cc=aou@eecs.berkeley.edu \
    --cc=conor@kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-riscv@lists.infradead.org \
    --cc=palmer@dabbelt.com \
    --cc=paul.walmsley@sifive.com \
    --cc=samuel.holland@sifive.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).