From: Chris Duncan <chris@zoso.dev>
Date: Tue, 14 Jan 2025 19:34:21 +0000 (-0800)
Subject: Replace some 16-bit scalar rotations with vector rotations.
X-Git-Url: https://zoso.dev/?a=commitdiff_plain;h=a521923b55b6a3028975d78c46fb5b934da01364;p=nano-pow.git

Replace some 16-bit scalar rotations with vector rotations.
---

diff --git a/src/shaders/compute.wgsl b/src/shaders/compute.wgsl
index 91d41be..3a887f5 100644
--- a/src/shaders/compute.wgsl
+++ b/src/shaders/compute.wgsl
@@ -192,10 +192,14 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	v1 = o1;
 
 	// d = rotr64(d ^ a, 16)
-	xor0 = v24 ^ v0;
-	xor1 = v25 ^ v1;
-	v24 = (xor0 >> 16u) ^ (xor1 << 16u);
-	v25 = (xor1 >> 16u) ^ (xor0 << 16u);
+	v_01.x = v0;
+	v_01.y = v1;
+	v_2425.x = v24;
+	v_2425.y = v25;
+	xor = v_2425 ^ v_01;
+	v_2425 = vec2((xor.x >> 16u) | (xor.y << 16u), (xor.y >> 16u) | (xor.x << 16u));
+	v24 = v_2425.x;
+	v25 = v_2425.y;
 
 	// c = c + d
 	v_1617.x = v16;
@@ -973,10 +977,14 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	// // skip since adding 0u does nothing
 
 	// d = rotr64(d ^ a, 16)
-	xor0 = v24 ^ v0;
-	xor1 = v25 ^ v1;
-	v24 = (xor0 >> 16u) ^ (xor1 << 16u);
-	v25 = (xor1 >> 16u) ^ (xor0 << 16u);
+	v_01.x = v0;
+	v_01.y = v1;
+	v_2425.x = v24;
+	v_2425.y = v25;
+	xor = v_2425 ^ v_01;
+	v_2425 = vec2((xor.x >> 16u) | (xor.y << 16u), (xor.y >> 16u) | (xor.x << 16u));
+	v24 = v_2425.x;
+	v25 = v_2425.y;
 
 	// c = c + d
 	v_1617.x = v16;
@@ -1757,10 +1765,14 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	// // skip since adding 0u does nothing
 
 	// d = rotr64(d ^ a, 16)
-	xor0 = v24 ^ v0;
-	xor1 = v25 ^ v1;
-	v24 = (xor0 >> 16u) ^ (xor1 << 16u);
-	v25 = (xor1 >> 16u) ^ (xor0 << 16u);
+	v_01.x = v0;
+	v_01.y = v1;
+	v_2425.x = v24;
+	v_2425.y = v25;
+	xor = v_2425 ^ v_01;
+	v_2425 = vec2((xor.x >> 16u) | (xor.y << 16u), (xor.y >> 16u) | (xor.x << 16u));
+	v24 = v_2425.x;
+	v25 = v_2425.y;
 
 	// c = c + d
 	v_1617.x = v16;
@@ -2532,10 +2544,14 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	// // skip since adding 0u does nothing
 
 	// d = rotr64(d ^ a, 16)
-	xor0 = v24 ^ v0;
-	xor1 = v25 ^ v1;
-	v24 = (xor0 >> 16u) ^ (xor1 << 16u);
-	v25 = (xor1 >> 16u) ^ (xor0 << 16u);
+	v_01.x = v0;
+	v_01.y = v1;
+	v_2425.x = v24;
+	v_2425.y = v25;
+	xor = v_2425 ^ v_01;
+	v_2425 = vec2((xor.x >> 16u) | (xor.y << 16u), (xor.y >> 16u) | (xor.x << 16u));
+	v24 = v_2425.x;
+	v25 = v_2425.y;
 
 	// c = c + d
 	v_1617.x = v16;
@@ -3322,10 +3338,14 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	v1 = o1;
 
 	// d = rotr64(d ^ a, 16)
-	xor0 = v24 ^ v0;
-	xor1 = v25 ^ v1;
-	v24 = (xor0 >> 16u) ^ (xor1 << 16u);
-	v25 = (xor1 >> 16u) ^ (xor0 << 16u);
+	v_01.x = v0;
+	v_01.y = v1;
+	v_2425.x = v24;
+	v_2425.y = v25;
+	xor = v_2425 ^ v_01;
+	v_2425 = vec2((xor.x >> 16u) | (xor.y << 16u), (xor.y >> 16u) | (xor.x << 16u));
+	v24 = v_2425.x;
+	v25 = v_2425.y;
 
 	// c = c + d
 	v_1617.x = v16;
@@ -4115,10 +4135,14 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	// // skip since adding 0u does nothing
 
 	// d = rotr64(d ^ a, 16)
-	xor0 = v24 ^ v0;
-	xor1 = v25 ^ v1;
-	v24 = (xor0 >> 16u) ^ (xor1 << 16u);
-	v25 = (xor1 >> 16u) ^ (xor0 << 16u);
+	v_01.x = v0;
+	v_01.y = v1;
+	v_2425.x = v24;
+	v_2425.y = v25;
+	xor = v_2425 ^ v_01;
+	v_2425 = vec2((xor.x >> 16u) | (xor.y << 16u), (xor.y >> 16u) | (xor.x << 16u));
+	v24 = v_2425.x;
+	v25 = v_2425.y;
 
 	// c = c + d
 	v_1617.x = v16;
@@ -4902,10 +4926,14 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	// // skip since adding 0u does nothing
 
 	// d = rotr64(d ^ a, 16)
-	xor0 = v24 ^ v0;
-	xor1 = v25 ^ v1;
-	v24 = (xor0 >> 16u) ^ (xor1 << 16u);
-	v25 = (xor1 >> 16u) ^ (xor0 << 16u);
+	v_01.x = v0;
+	v_01.y = v1;
+	v_2425.x = v24;
+	v_2425.y = v25;
+	xor = v_2425 ^ v_01;
+	v_2425 = vec2((xor.x >> 16u) | (xor.y << 16u), (xor.y >> 16u) | (xor.x << 16u));
+	v24 = v_2425.x;
+	v25 = v_2425.y;
 
 	// c = c + d
 	v_1617.x = v16;
@@ -5686,10 +5714,14 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	// // skip since adding 0u does nothing
 
 	// d = rotr64(d ^ a, 16)
-	xor0 = v24 ^ v0;
-	xor1 = v25 ^ v1;
-	v24 = (xor0 >> 16u) ^ (xor1 << 16u);
-	v25 = (xor1 >> 16u) ^ (xor0 << 16u);
+	v_01.x = v0;
+	v_01.y = v1;
+	v_2425.x = v24;
+	v_2425.y = v25;
+	xor = v_2425 ^ v_01;
+	v_2425 = vec2((xor.x >> 16u) | (xor.y << 16u), (xor.y >> 16u) | (xor.x << 16u));
+	v24 = v_2425.x;
+	v25 = v_2425.y;
 
 	// c = c + d
 	v_1617.x = v16;
@@ -6470,10 +6502,14 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	// // skip since adding 0u does nothing
 
 	// d = rotr64(d ^ a, 16)
-	xor0 = v24 ^ v0;
-	xor1 = v25 ^ v1;
-	v24 = (xor0 >> 16u) ^ (xor1 << 16u);
-	v25 = (xor1 >> 16u) ^ (xor0 << 16u);
+	v_01.x = v0;
+	v_01.y = v1;
+	v_2425.x = v24;
+	v_2425.y = v25;
+	xor = v_2425 ^ v_01;
+	v_2425 = vec2((xor.x >> 16u) | (xor.y << 16u), (xor.y >> 16u) | (xor.x << 16u));
+	v24 = v_2425.x;
+	v25 = v_2425.y;
 
 	// c = c + d
 	v_1617.x = v16;
@@ -7260,10 +7296,14 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	v1 = o1;
 
 	// d = rotr64(d ^ a, 16)
-	xor0 = v24 ^ v0;
-	xor1 = v25 ^ v1;
-	v24 = (xor0 >> 16u) ^ (xor1 << 16u);
-	v25 = (xor1 >> 16u) ^ (xor0 << 16u);
+	v_01.x = v0;
+	v_01.y = v1;
+	v_2425.x = v24;
+	v_2425.y = v25;
+	xor = v_2425 ^ v_01;
+	v_2425 = vec2((xor.x >> 16u) | (xor.y << 16u), (xor.y >> 16u) | (xor.x << 16u));
+	v24 = v_2425.x;
+	v25 = v_2425.y;
 
 	// c = c + d
 	v_1617.x = v16;
@@ -8050,10 +8090,14 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	v1 = o1;
 
 	// d = rotr64(d ^ a, 16)
-	xor0 = v24 ^ v0;
-	xor1 = v25 ^ v1;
-	v24 = (xor0 >> 16u) ^ (xor1 << 16u);
-	v25 = (xor1 >> 16u) ^ (xor0 << 16u);
+	v_01.x = v0;
+	v_01.y = v1;
+	v_2425.x = v24;
+	v_2425.y = v25;
+	xor = v_2425 ^ v_01;
+	v_2425 = vec2((xor.x >> 16u) | (xor.y << 16u), (xor.y >> 16u) | (xor.x << 16u));
+	v24 = v_2425.x;
+	v25 = v_2425.y;
 
 	// c = c + d
 	v_1617.x = v16;
@@ -8831,10 +8875,14 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	// // skip since adding 0u does nothing
 
 	// d = rotr64(d ^ a, 16)
-	xor0 = v24 ^ v0;
-	xor1 = v25 ^ v1;
-	v24 = (xor0 >> 16u) ^ (xor1 << 16u);
-	v25 = (xor1 >> 16u) ^ (xor0 << 16u);
+	v_01.x = v0;
+	v_01.y = v1;
+	v_2425.x = v24;
+	v_2425.y = v25;
+	xor = v_2425 ^ v_01;
+	v_2425 = vec2((xor.x >> 16u) | (xor.y << 16u), (xor.y >> 16u) | (xor.x << 16u));
+	v24 = v_2425.x;
+	v25 = v_2425.y;
 
 	// c = c + d
 	v_1617.x = v16;