From: Konstantinos Margaritis <konstantinos.margaritis@vectorcamp.gr>
Date: Wed, 21 Sep 2022 13:06:13 +0000 (+0000)
Subject: First form of fully working SVP64 version
X-Git-Url: https://git.libre-soc.org/?a=commitdiff_plain;h=f588d073b576aa30c9cad76b2e77160439ad2da5;p=openpower-isa.git

First form of fully working SVP64 version
---

diff --git a/media/video/libvpx/variancefuncs_svp64.s b/media/video/libvpx/variancefuncs_svp64.s
deleted file mode 100644
index 86541450..00000000
--- a/media/video/libvpx/variancefuncs_svp64.s
+++ /dev/null
@@ -1,39 +0,0 @@
-.set in, 3
-.set vin, 20
-.set sum, 6
-.set ctr, 9
-
-
-	.machine libresoc
-	.file	"variancefuncs_svp64.c"
-	.text
-	.abiversion 2
-	.section	".text"
-	.align 2
-	.globl vpx_get_mb_ss_svp64_real
-	.type	vpx_get_mb_ss_svp64_real, @function
-vpx_get_mb_ss_svp64_real:
-.LFB0:
-	.cfi_startproc
-	addi 10, in ,-2
-	li in, 0
-	li sum, 0
-	li ctr, 4
-	mtctr ctr
-	setvl 0,0,64,0,1,1			# Set VL to 64 elements
-.L2:
-	# Load 64 ints from (in)
-	sv.lha	 	*vin, 0(in)
-	# equivalent to: for (i = 0; i < 64; i++) sum += in[i] * in[i];
-	sv.maddld 	sum, *vin, *vin, sum
-	addi in, in, 16
-	rldicl in,ctr,0,32
-	bdnz .L2
-	blr
-	.long 0
-	.byte 0,0,0,0,0,0,0,0
-	.cfi_endproc
-.LFE0:
-	.size	vpx_get_mb_ss_svp64_real,.-vpx_get_mb_ss_svp64_real
-	.ident	"GCC: (Debian 8.3.0-6) 8.3.0"
-	.section	.note.GNU-stack,"",@progbits
diff --git a/media/video/libvpx/vpx_get_mb_ss_svp64_real.s b/media/video/libvpx/vpx_get_mb_ss_svp64_real.s
new file mode 100644
index 00000000..c8cae746
--- /dev/null
+++ b/media/video/libvpx/vpx_get_mb_ss_svp64_real.s
@@ -0,0 +1,42 @@
+.set in, 3
+.set src, 10
+.set prod, 50
+.set sum, 4
+.set ctr, 9
+
+
+	.machine libresoc
+	.file	"variancefuncs_svp64.c"
+	.text
+	.abiversion 2
+	.section	".text"
+	.align 2
+	.globl vpx_get_mb_ss_svp64_real
+	.type	vpx_get_mb_ss_svp64_real, @function
+vpx_get_mb_ss_svp64_real:
+.LFB0:
+	.cfi_startproc
+	li sum, 0
+	li ctr, 8
+	mtctr ctr
+	setvl 0,0,32,0,1,1			# Set VL to 64 elements
+.L2:
+	# Load 32 ints from (in)
+	sv.lha	 	*src, 0(in)
+	# equivalent to: for (i = 0; i < 32; i++) vprod[i] = src[i] * src[i];
+	sv.mulld 	*prod, *src, *src
+	# equivalent to: for (i = 0; i < 32; i++) sum += prod[i];
+	sv.add/mr	sum, *prod, sum
+	addi in, in, 64
+#	rldicl in,ctr,0,32
+	bdnz .L2
+	li in, 0
+	addi in, sum, 0
+	blr
+	.long 0
+	.byte 0,0,0,0,0,0,0,0
+	.cfi_endproc
+.LFE0:
+	.size	vpx_get_mb_ss_svp64_real,.-vpx_get_mb_ss_svp64_real
+	.ident	"GCC: (Debian 8.3.0-6) 8.3.0"
+	.section	.note.GNU-stack,"",@progbits