linux-hwmon.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Guenter Roeck <linux@roeck-us.net>
To: "Grönke, Christian" <C.Groenke@infodas.de>
Cc: "linux-hwmon@vger.kernel.org" <linux-hwmon@vger.kernel.org>
Subject: Re: PMBus driver for FSP/3Y Power device with non-standard VOUT values (LINEAR11 vs LINEAR16)
Date: Wed, 13 Mar 2019 20:18:49 -0700	[thread overview]
Message-ID: <20190314031849.GA19793@roeck-us.net> (raw)
In-Reply-To: <7c2e3520f1d24e51b7002c8d0138f00a@infodas.de>

Hi Christian,

On Wed, Mar 13, 2019 at 05:35:38PM +0000, Grönke, Christian wrote:
[ ... ]
> > 
> > Our last e-mails overlapped; can you explore what it would take to add
> > support for ieee754 half precision floating point ?  We would need a new
> > set of conversion functions in the core (ie pmbus_reg2data_ieee754 and
> > pmbus_data2reg_ieee754), and your driver would have to implement the
> > linear11 <-> ieee754 conversion. That may be a bit more work, but it
> > would also be cleaner, and it should not affect precision.
> 
> I give it a look tomorrow. If this presents a proper and clean way, 
> to avoid adding the hack in the generic code it might be worth it.
> 

Please use the attached patch as starting point. Obviously I could not do much
testing, but unit testing returned reasonable results. Note that we don't have
to follow the pmbus spec to the letter - it is sufficient to declare the VOUT
values to be in ieee754 format for your driver.

Guenter

---
From 2f80cad52914c674afc5a80a84f4756e1b12d803 Mon Sep 17 00:00:00 2001
From: Guenter Roeck <linux@roeck-us.net>
Date: Wed, 13 Mar 2019 14:36:28 -0700
Subject: [PATCH] hwmon: (pmbus) Add IEEE 754 half precision support to PMBus
 core

PMBus v1.3.1 adds IEEE 754 half precision as additional data format.
Add support for it.

Signed-off-by: Guenter Roeck <linux@roeck-us.net>
---
 drivers/hwmon/pmbus/pmbus.h      |   2 +-
 drivers/hwmon/pmbus/pmbus_core.c | 135 +++++++++++++++++++++++++++++++++++++--
 2 files changed, 132 insertions(+), 5 deletions(-)

diff --git a/drivers/hwmon/pmbus/pmbus.h b/drivers/hwmon/pmbus/pmbus.h
index 1d24397d36ec..6477eb433790 100644
--- a/drivers/hwmon/pmbus/pmbus.h
+++ b/drivers/hwmon/pmbus/pmbus.h
@@ -374,7 +374,7 @@ enum pmbus_sensor_classes {
 
 #define PMBUS_PAGE_VIRTUAL	BIT(31)
 
-enum pmbus_data_format { linear = 0, direct, vid };
+enum pmbus_data_format { linear = 0, ieee754, direct, vid };
 enum vrm_version { vr11 = 0, vr12, vr13 };
 
 struct pmbus_driver_info {
diff --git a/drivers/hwmon/pmbus/pmbus_core.c b/drivers/hwmon/pmbus/pmbus_core.c
index 2e2b5851139c..e828160a1793 100644
--- a/drivers/hwmon/pmbus/pmbus_core.c
+++ b/drivers/hwmon/pmbus/pmbus_core.c
@@ -619,6 +619,66 @@ static struct pmbus_data *pmbus_update_device(struct device *dev)
 }
 
 /*
+ * Convert ieee754 sensor values to milli- or micro-units
+ * depending on sensor type.
+ *
+ * ieee754 data format:
+ *	bit 15:		sign
+ *	bit 10..14:	exponent
+ *	bit 0..9:	mantissa
+ * exponent=0:
+ *	v=(−1)^signbit * 2^(−14) * 0.significantbits
+ * exponent=1..30:
+ *	v=(−1)^signbit * 2^(exponent - 15) * 1.significantbits
+ * exponent=31:
+ *	v=NaN
+ *
+ * Add the number mantissa bits into the calculations for simplicity.
+ * To do that, add '10' to the exponent. By doing that, we can just add
+ * 0x400 to normal values and get the expected result.
+ */
+static long pmbus_reg2data_ieee754(struct pmbus_data *data,
+				   struct pmbus_sensor *sensor)
+{
+	int exponent;
+	bool sign;
+	long val;
+
+	/* only support half precision for now */
+	sign = sensor->data & 0x8000;
+	exponent = (sensor->data >> 10) & 0x1f;
+	val = sensor->data & 0x3ff;
+
+	if (exponent == 0) {			/* subnormal */
+		exponent = -(14 + 10);
+	} else if (exponent ==  0x1f) {		/* NaN, convert to min/max */
+		exponent = 0;
+		val = 65504;
+	} else {
+		exponent -= (15 + 10);		/* normal */
+		val |= 0x400;
+	}
+
+	/* scale result to milli-units for all sensors except fans */
+	if (sensor->class != PSC_FAN)
+		val = val * 1000L;
+
+	/* scale result to micro-units for power sensors */
+	if (sensor->class == PSC_POWER)
+		val = val * 1000L;
+
+	if (exponent >= 0)
+		val <<= exponent;
+	else
+		val >>= -exponent;
+
+	if (sign)
+		val = -val;
+
+	return val;
+}
+
+/*
  * Convert linear sensor values to milli- or micro-units
  * depending on sensor type.
  */
@@ -740,6 +800,9 @@ static long pmbus_reg2data(struct pmbus_data *data, struct pmbus_sensor *sensor)
 	case vid:
 		val = pmbus_reg2data_vid(data, sensor);
 		break;
+	case ieee754:
+		val = pmbus_reg2data_ieee754(data, sensor);
+		break;
 	case linear:
 	default:
 		val = pmbus_reg2data_linear(data, sensor);
@@ -748,8 +811,65 @@ static long pmbus_reg2data(struct pmbus_data *data, struct pmbus_sensor *sensor)
 	return val;
 }
 
-#define MAX_MANTISSA	(1023 * 1000)
-#define MIN_MANTISSA	(511 * 1000)
+#define MAX_IEEE_MANTISSA	(0x7ff * 1000)
+#define MIN_IEEE_MANTISSA	(0x400 * 1000)
+
+static u16 pmbus_data2reg_ieee754(struct pmbus_data *data,
+				  struct pmbus_sensor *sensor, long val)
+{
+	u16 exponent = (15 + 10);
+	long mantissa;
+	u16 sign = 0;
+
+	/* simple case */
+	if (val == 0)
+		return 0;
+
+	if (val < 0) {
+		sign = 0x8000;
+		val = -val;
+	}
+
+	/* Power is in uW. Convert to mW before converting. */
+	if (sensor->class == PSC_POWER)
+		val = DIV_ROUND_CLOSEST(val, 1000L);
+
+	/*
+	 * For simplicity, convert fan data to milli-units
+	 * before calculating the exponent.
+	 */
+	if (sensor->class == PSC_FAN)
+		val = val * 1000;
+
+	/* Reduce large mantissa until it fits into 10 bit */
+	while (val > MAX_IEEE_MANTISSA && exponent < 30) {
+		exponent++;
+		val >>= 1;
+	}
+	/*
+	 * Increase small mantissa to generate valid 'normal'
+	 * number
+	 */
+	while (val < MIN_IEEE_MANTISSA && exponent > 1) {
+		exponent--;
+		val <<= 1;
+	}
+
+	/* Convert mantissa from milli-units to units */
+	mantissa = DIV_ROUND_CLOSEST(val, 1000);
+
+	/* Ensure that resulting number is within range */
+	if (mantissa > 0x7ff)
+		mantissa = 0x7ff;
+	else if (mantissa < 0x400)
+		mantissa = 0x400;
+
+	/* Convert to sign, 5 bit exponent, 10 bit mantissa */
+	return sign | (mantissa & 0x3ff) | ((exponent << 10) & 0x7c00);
+}
+
+#define MAX_LIN_MANTISSA	(1023 * 1000)
+#define MIN_LIN_MANTISSA	(511 * 1000)
 
 static u16 pmbus_data2reg_linear(struct pmbus_data *data,
 				 struct pmbus_sensor *sensor, long val)
@@ -795,12 +915,12 @@ static u16 pmbus_data2reg_linear(struct pmbus_data *data,
 		val = val * 1000;
 
 	/* Reduce large mantissa until it fits into 10 bit */
-	while (val >= MAX_MANTISSA && exponent < 15) {
+	while (val >= MAX_LIN_MANTISSA && exponent < 15) {
 		exponent++;
 		val >>= 1;
 	}
 	/* Increase small mantissa to improve precision */
-	while (val < MIN_MANTISSA && exponent > -15) {
+	while (val < MIN_LIN_MANTISSA && exponent > -15) {
 		exponent--;
 		val <<= 1;
 	}
@@ -878,6 +998,9 @@ static u16 pmbus_data2reg(struct pmbus_data *data,
 	case vid:
 		regval = pmbus_data2reg_vid(data, sensor, val);
 		break;
+	case ieee754:
+		regval = pmbus_data2reg_ieee754(data, sensor, val);
+		break;
 	case linear:
 	default:
 		regval = pmbus_data2reg_linear(data, sensor, val);
@@ -1967,6 +2090,10 @@ static int pmbus_identify_common(struct i2c_client *client,
 			if (data->info->format[PSC_VOLTAGE_OUT] != direct)
 				return -ENODEV;
 			break;
+		case 3:	/* ieee 754 half precision */
+			if (data->info->format[PSC_VOLTAGE_OUT] != ieee754)
+				return -ENODEV;
+			break;
 		default:
 			return -ENODEV;
 		}
-- 
2.7.4


  reply	other threads:[~2019-03-14  3:18 UTC|newest]

Thread overview: 14+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2019-03-13 12:31 PMBus driver for FSP/3Y Power device with non-standard VOUT values (LINEAR11 vs LINEAR16) Grönke, Christian
2019-03-13 15:19 ` Guenter Roeck
2019-03-13 16:20   ` AW: " Grönke, Christian
2019-03-13 16:30     ` Guenter Roeck
2019-03-13 17:35       ` AW: " Grönke, Christian
2019-03-14  3:18         ` Guenter Roeck [this message]
2019-03-14 16:08           ` Grönke, Christian
2019-03-14 16:22             ` Guenter Roeck
2019-03-14 16:53             ` Guenter Roeck
2019-03-15 10:19               ` AW: " Grönke, Christian
2019-03-15 18:29                 ` Guenter Roeck
2019-03-15 19:38                   ` Guenter Roeck
2019-03-16 15:27                     ` Guenter Roeck
2019-03-13 16:21   ` Guenter Roeck

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=20190314031849.GA19793@roeck-us.net \
    --to=linux@roeck-us.net \
    --cc=C.Groenke@infodas.de \
    --cc=linux-hwmon@vger.kernel.org \
    /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).