From: Chris Duncan <chris@zoso.dev>
Date: Tue, 14 Jan 2025 06:04:36 +0000 (-0800)
Subject: Stage final fully reduced u64 addition operation using vectors. This will be the... 
X-Git-Url: https://zoso.dev/?a=commitdiff_plain;h=9c790daed644acfe9f9d1eb43852c91cf96b4ef1;p=nano-pow.git

Stage final fully reduced u64 addition operation using vectors. This will be the technique going forward; the assignments to individual scalars after the vector sum are purely placeholders until the entire algorithm can be converted.
---

diff --git a/src/shaders/compute.wgsl b/src/shaders/compute.wgsl
index 16f2af1..17cfdd6 100644
--- a/src/shaders/compute.wgsl
+++ b/src/shaders/compute.wgsl
@@ -57,6 +57,10 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	* v[28u] = ~v[28u];
 	* v[29u] = ~v[29u];
 	*/
+	var v_01: vec2<u32> = vec2(0xF2BDC900u, 0x6A09E667u);
+	var v_23: vec2<u32> = vec2(0x84CAA73Bu, 0xBB67AE85u);
+	var v_89: vec2<u32> = vec2(0xADE682D1u, 0x510E527Fu);
+	var v_1011: vec2<u32> = vec2(0x2B3E6C1Fu, 0x9B05688Cu);
 	var v0: u32 = 0xF2BDC900u;
 	var v1: u32 = 0x6A09E667u;
 	var v2: u32 = 0x84CAA73Bu;
@@ -114,9 +118,14 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	*/
 
 	// a = a + b
-	v1 = v1 + v9;
-	v1 = v1 + select(0u, 1u, v0 + v8 < v0);
-	v0 = v0 + v8;
+	// v1 = v1 + v9 + select(0u, 1u, v0 + v8 < v0);
+	// v0 = v0 + v8;
+	v_01 = v_01 + v_89 + select(vec2<u32>(0u), vec2<u32>(0u, 1u), v_01.x + v_89.x < v_01.x);
+	v0 = v_01.x;
+	v1 = v_01.y;
+	v8 = v_89.x;
+	v9 = v_89.y;
+
 
 	// a = a + m[sigma[r][2*i+0]]
 	o0 = v0 + m0;
@@ -184,11 +193,18 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	*/
 
 	// a = a + b
-	o0 = v2 + v10;
-	o1 = v3 + v11;
-	o1 = o1 + select(0u, 1u, o0 < v2);
-	v2 = o0;
-	v3 = o1;
+	v_23 = v_23 + v_1011 + select(vec2<u32>(0u), vec2<u32>(0u, 1u), v_23.x + v_1011.x < v_23.x);
+	v2 = v_23.x;
+	v3 = v_23.y;
+	v10 = v_1011.x;
+	v11 = v_1011.y;
+
+	// o0 = v2 + v10;
+	// o1 = v3 + v11;
+	// o1 = o1 + select(0u, 1u, o0 < v2);
+	// v2 = o0;
+	// v3 = o1;
+
 
 	// a = a + m[sigma[r][2*i+0]]
 	o0 = v2 + m4;