From d71f3c683c86840f68a9ae94489aef9437bc536d Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=C5=81ukasz=20Magiera?= <magik6k@users.noreply.github.com>
Date: Wed, 16 Oct 2024 14:37:18 +0200
Subject: [PATCH] experiment: GPU Overprovisioning (#262)

* experiment: GPU Overprovisioning

* make gen
---
 harmony/resources/getGPU.go | 17 +++++++++++++++--
 lib/ffiselect/ffiselect.go  |  9 ++++++---
 2 files changed, 21 insertions(+), 5 deletions(-)

diff --git a/harmony/resources/getGPU.go b/harmony/resources/getGPU.go
index 62d5c091e..a9a207125 100644
--- a/harmony/resources/getGPU.go
+++ b/harmony/resources/getGPU.go
@@ -11,6 +11,19 @@ import (
 	ffi "github.com/filecoin-project/filecoin-ffi"
 )
 
+var GpuOverprovisionFactor = 1
+
+func init() {
+	if nstr := os.Getenv("HARMONY_GPU_OVERPROVISION_FACTOR"); nstr != "" {
+		n, err := strconv.Atoi(nstr)
+		if err != nil {
+			logger.Errorf("parsing HARMONY_GPU_OVERPROVISION_FACTOR failed: %+v", err)
+		} else {
+			GpuOverprovisionFactor = n
+		}
+	}
+}
+
 func getGPUDevices() float64 { // GPU boolean
 	if nstr := os.Getenv("HARMONY_OVERRIDE_GPUS"); nstr != "" {
 		n, err := strconv.ParseFloat(nstr, 64)
@@ -22,13 +35,13 @@ func getGPUDevices() float64 { // GPU boolean
 	}
 
 	gpus, err := ffi.GetGPUDevices()
-	logger.Infow("GPUs", "list", gpus)
+	logger.Infow("GPUs", "list", gpus, "overprovision_factor", GpuOverprovisionFactor)
 	if err != nil {
 		logger.Errorf("getting gpu devices failed: %+v", err)
 	}
 	all := strings.ToLower(strings.Join(gpus, ","))
 	if len(gpus) > 1 || strings.Contains(all, "ati") || strings.Contains(all, "nvidia") {
-		return float64(len(gpus))
+		return float64(len(gpus) * GpuOverprovisionFactor)
 	}
 	return 0
 }
diff --git a/lib/ffiselect/ffiselect.go b/lib/ffiselect/ffiselect.go
index 5122c8453..d0267ab73 100644
--- a/lib/ffiselect/ffiselect.go
+++ b/lib/ffiselect/ffiselect.go
@@ -19,6 +19,7 @@ import (
 	"github.com/filecoin-project/go-state-types/proof"
 
 	"github.com/filecoin-project/curio/build"
+	"github.com/filecoin-project/curio/harmony/resources"
 	"github.com/filecoin-project/curio/lib/storiface"
 )
 
@@ -45,9 +46,11 @@ func init() {
 		ch = make(chan string, 1)
 		ch <- "0"
 	} else {
-		ch = make(chan string, len(devices))
-		for i := 0; i < len(devices); i++ {
-			ch <- strconv.Itoa(i)
+		nSlots := len(devices) * resources.GpuOverprovisionFactor
+
+		ch = make(chan string, nSlots)
+		for i := 0; i < nSlots; i++ {
+			ch <- strconv.Itoa(i / resources.GpuOverprovisionFactor)
 		}
 	}
 }