From 255fdbdfbe0788b977c0e0b12d328434103ea3d8 Mon Sep 17 00:00:00 2001
From: Chris Duncan <chris@zoso.dev>
Date: Tue, 14 Jan 2025 08:26:17 -0800
Subject: [PATCH] Repalce more scalar addition with vector addition.

---
 src/shaders/compute.wgsl | 336 +++++++++++++++++++++++++--------------
 1 file changed, 216 insertions(+), 120 deletions(-)

diff --git a/src/shaders/compute.wgsl b/src/shaders/compute.wgsl
index 3162f0a..ff73ed2 100644
--- a/src/shaders/compute.wgsl
+++ b/src/shaders/compute.wgsl
@@ -523,11 +523,15 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	v31 = xor0;
 
 	// c = c + d
-	o0 = v20 + v30;
-	o1 = v21 + v31;
-	o1 = o1 + select(0u, 1u, o0 < v20);
-	v20 = o0;
-	v21 = o1;
+	v_2021.x = v20;
+	v_2021.y = v21;
+	v_3031.x = v30;
+	v_3031.y = v31;
+	v_2021 = v_2021 + v_3031 + select(vec2<u32>(0u), vec2<u32>(0u, 1u), v_2021.x + v_3031.x < v_2021.x);
+	v20 = v_2021.x;
+	v21 = v_2021.y;
+	v30 = v_3031.x;
+	v31 = v_3031.y;
 
 	// b = rotr64(b ^ c, 24)
 	xor0 = v10 ^ v20;
@@ -563,11 +567,15 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	v31 = (xor1 >> 16u) ^ (xor0 << 16u);
 
 	// c = c + d
-	o0 = v20 + v30;
-	o1 = v21 + v31;
-	o1 = o1 + select(0u, 1u, o0 < v20);
-	v20 = o0;
-	v21 = o1;
+	v_2021.x = v20;
+	v_2021.y = v21;
+	v_3031.x = v30;
+	v_3031.y = v31;
+	v_2021 = v_2021 + v_3031 + select(vec2<u32>(0u), vec2<u32>(0u, 1u), v_2021.x + v_3031.x < v_2021.x);
+	v20 = v_2021.x;
+	v21 = v_2021.y;
+	v30 = v_3031.x;
+	v31 = v_3031.y;
 
 	// b = rotr64(b ^ c, 63)
 	xor0 = v10 ^ v20;
@@ -1233,11 +1241,15 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	v31 = xor0;
 
 	// c = c + d
-	o0 = v20 + v30;
-	o1 = v21 + v31;
-	o1 = o1 + select(0u, 1u, o0 < v20);
-	v20 = o0;
-	v21 = o1;
+	v_2021.x = v20;
+	v_2021.y = v21;
+	v_3031.x = v30;
+	v_3031.y = v31;
+	v_2021 = v_2021 + v_3031 + select(vec2<u32>(0u), vec2<u32>(0u, 1u), v_2021.x + v_3031.x < v_2021.x);
+	v20 = v_2021.x;
+	v21 = v_2021.y;
+	v30 = v_3031.x;
+	v31 = v_3031.y;
 
 	// b = rotr64(b ^ c, 24)
 	xor0 = v10 ^ v20;
@@ -1273,11 +1285,15 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	v31 = (xor1 >> 16u) ^ (xor0 << 16u);
 
 	// c = c + d
-	o0 = v20 + v30;
-	o1 = v21 + v31;
-	o1 = o1 + select(0u, 1u, o0 < v20);
-	v20 = o0;
-	v21 = o1;
+	v_2021.x = v20;
+	v_2021.y = v21;
+	v_3031.x = v30;
+	v_3031.y = v31;
+	v_2021 = v_2021 + v_3031 + select(vec2<u32>(0u), vec2<u32>(0u, 1u), v_2021.x + v_3031.x < v_2021.x);
+	v20 = v_2021.x;
+	v21 = v_2021.y;
+	v30 = v_3031.x;
+	v31 = v_3031.y;
 
 	// b = rotr64(b ^ c, 63)
 	xor0 = v10 ^ v20;
@@ -1934,11 +1950,15 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	v31 = xor0;
 
 	// c = c + d
-	o0 = v20 + v30;
-	o1 = v21 + v31;
-	o1 = o1 + select(0u, 1u, o0 < v20);
-	v20 = o0;
-	v21 = o1;
+	v_2021.x = v20;
+	v_2021.y = v21;
+	v_3031.x = v30;
+	v_3031.y = v31;
+	v_2021 = v_2021 + v_3031 + select(vec2<u32>(0u), vec2<u32>(0u, 1u), v_2021.x + v_3031.x < v_2021.x);
+	v20 = v_2021.x;
+	v21 = v_2021.y;
+	v30 = v_3031.x;
+	v31 = v_3031.y;
 
 	// b = rotr64(b ^ c, 24)
 	xor0 = v10 ^ v20;
@@ -1974,11 +1994,15 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	v31 = (xor1 >> 16u) ^ (xor0 << 16u);
 
 	// c = c + d
-	o0 = v20 + v30;
-	o1 = v21 + v31;
-	o1 = o1 + select(0u, 1u, o0 < v20);
-	v20 = o0;
-	v21 = o1;
+	v_2021.x = v20;
+	v_2021.y = v21;
+	v_3031.x = v30;
+	v_3031.y = v31;
+	v_2021 = v_2021 + v_3031 + select(vec2<u32>(0u), vec2<u32>(0u, 1u), v_2021.x + v_3031.x < v_2021.x);
+	v20 = v_2021.x;
+	v21 = v_2021.y;
+	v30 = v_3031.x;
+	v31 = v_3031.y;
 
 	// b = rotr64(b ^ c, 63)
 	xor0 = v10 ^ v20;
@@ -2632,11 +2656,15 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	v31 = xor0;
 
 	// c = c + d
-	o0 = v20 + v30;
-	o1 = v21 + v31;
-	o1 = o1 + select(0u, 1u, o0 < v20);
-	v20 = o0;
-	v21 = o1;
+	v_2021.x = v20;
+	v_2021.y = v21;
+	v_3031.x = v30;
+	v_3031.y = v31;
+	v_2021 = v_2021 + v_3031 + select(vec2<u32>(0u), vec2<u32>(0u, 1u), v_2021.x + v_3031.x < v_2021.x);
+	v20 = v_2021.x;
+	v21 = v_2021.y;
+	v30 = v_3031.x;
+	v31 = v_3031.y;
 
 	// b = rotr64(b ^ c, 24)
 	xor0 = v10 ^ v20;
@@ -2672,11 +2700,15 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	v31 = (xor1 >> 16u) ^ (xor0 << 16u);
 
 	// c = c + d
-	o0 = v20 + v30;
-	o1 = v21 + v31;
-	o1 = o1 + select(0u, 1u, o0 < v20);
-	v20 = o0;
-	v21 = o1;
+	v_2021.x = v20;
+	v_2021.y = v21;
+	v_3031.x = v30;
+	v_3031.y = v31;
+	v_2021 = v_2021 + v_3031 + select(vec2<u32>(0u), vec2<u32>(0u, 1u), v_2021.x + v_3031.x < v_2021.x);
+	v20 = v_2021.x;
+	v21 = v_2021.y;
+	v30 = v_3031.x;
+	v31 = v_3031.y;
 
 	// b = rotr64(b ^ c, 63)
 	xor0 = v10 ^ v20;
@@ -3333,11 +3365,15 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	v31 = xor0;
 
 	// c = c + d
-	o0 = v20 + v30;
-	o1 = v21 + v31;
-	o1 = o1 + select(0u, 1u, o0 < v20);
-	v20 = o0;
-	v21 = o1;
+	v_2021.x = v20;
+	v_2021.y = v21;
+	v_3031.x = v30;
+	v_3031.y = v31;
+	v_2021 = v_2021 + v_3031 + select(vec2<u32>(0u), vec2<u32>(0u, 1u), v_2021.x + v_3031.x < v_2021.x);
+	v20 = v_2021.x;
+	v21 = v_2021.y;
+	v30 = v_3031.x;
+	v31 = v_3031.y;
 
 	// b = rotr64(b ^ c, 24)
 	xor0 = v10 ^ v20;
@@ -3370,11 +3406,15 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	v31 = (xor1 >> 16u) ^ (xor0 << 16u);
 
 	// c = c + d
-	o0 = v20 + v30;
-	o1 = v21 + v31;
-	o1 = o1 + select(0u, 1u, o0 < v20);
-	v20 = o0;
-	v21 = o1;
+	v_2021.x = v20;
+	v_2021.y = v21;
+	v_3031.x = v30;
+	v_3031.y = v31;
+	v_2021 = v_2021 + v_3031 + select(vec2<u32>(0u), vec2<u32>(0u, 1u), v_2021.x + v_3031.x < v_2021.x);
+	v20 = v_2021.x;
+	v21 = v_2021.y;
+	v30 = v_3031.x;
+	v31 = v_3031.y;
 
 	// b = rotr64(b ^ c, 63)
 	xor0 = v10 ^ v20;
@@ -4031,11 +4071,15 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	v31 = xor0;
 
 	// c = c + d
-	o0 = v20 + v30;
-	o1 = v21 + v31;
-	o1 = o1 + select(0u, 1u, o0 < v20);
-	v20 = o0;
-	v21 = o1;
+	v_2021.x = v20;
+	v_2021.y = v21;
+	v_3031.x = v30;
+	v_3031.y = v31;
+	v_2021 = v_2021 + v_3031 + select(vec2<u32>(0u), vec2<u32>(0u, 1u), v_2021.x + v_3031.x < v_2021.x);
+	v20 = v_2021.x;
+	v21 = v_2021.y;
+	v30 = v_3031.x;
+	v31 = v_3031.y;
 
 	// b = rotr64(b ^ c, 24)
 	xor0 = v10 ^ v20;
@@ -4071,11 +4115,15 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	v31 = (xor1 >> 16u) ^ (xor0 << 16u);
 
 	// c = c + d
-	o0 = v20 + v30;
-	o1 = v21 + v31;
-	o1 = o1 + select(0u, 1u, o0 < v20);
-	v20 = o0;
-	v21 = o1;
+	v_2021.x = v20;
+	v_2021.y = v21;
+	v_3031.x = v30;
+	v_3031.y = v31;
+	v_2021 = v_2021 + v_3031 + select(vec2<u32>(0u), vec2<u32>(0u, 1u), v_2021.x + v_3031.x < v_2021.x);
+	v20 = v_2021.x;
+	v21 = v_2021.y;
+	v30 = v_3031.x;
+	v31 = v_3031.y;
 
 	// b = rotr64(b ^ c, 63)
 	xor0 = v10 ^ v20;
@@ -4735,11 +4783,15 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	v31 = xor0;
 
 	// c = c + d
-	o0 = v20 + v30;
-	o1 = v21 + v31;
-	o1 = o1 + select(0u, 1u, o0 < v20);
-	v20 = o0;
-	v21 = o1;
+	v_2021.x = v20;
+	v_2021.y = v21;
+	v_3031.x = v30;
+	v_3031.y = v31;
+	v_2021 = v_2021 + v_3031 + select(vec2<u32>(0u), vec2<u32>(0u, 1u), v_2021.x + v_3031.x < v_2021.x);
+	v20 = v_2021.x;
+	v21 = v_2021.y;
+	v30 = v_3031.x;
+	v31 = v_3031.y;
 
 	// b = rotr64(b ^ c, 24)
 	xor0 = v10 ^ v20;
@@ -4775,11 +4827,15 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	v31 = (xor1 >> 16u) ^ (xor0 << 16u);
 
 	// c = c + d
-	o0 = v20 + v30;
-	o1 = v21 + v31;
-	o1 = o1 + select(0u, 1u, o0 < v20);
-	v20 = o0;
-	v21 = o1;
+	v_2021.x = v20;
+	v_2021.y = v21;
+	v_3031.x = v30;
+	v_3031.y = v31;
+	v_2021 = v_2021 + v_3031 + select(vec2<u32>(0u), vec2<u32>(0u, 1u), v_2021.x + v_3031.x < v_2021.x);
+	v20 = v_2021.x;
+	v21 = v_2021.y;
+	v30 = v_3031.x;
+	v31 = v_3031.y;
 
 	// b = rotr64(b ^ c, 63)
 	xor0 = v10 ^ v20;
@@ -5439,11 +5495,15 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	v31 = xor0;
 
 	// c = c + d
-	o0 = v20 + v30;
-	o1 = v21 + v31;
-	o1 = o1 + select(0u, 1u, o0 < v20);
-	v20 = o0;
-	v21 = o1;
+	v_2021.x = v20;
+	v_2021.y = v21;
+	v_3031.x = v30;
+	v_3031.y = v31;
+	v_2021 = v_2021 + v_3031 + select(vec2<u32>(0u), vec2<u32>(0u, 1u), v_2021.x + v_3031.x < v_2021.x);
+	v20 = v_2021.x;
+	v21 = v_2021.y;
+	v30 = v_3031.x;
+	v31 = v_3031.y;
 
 	// b = rotr64(b ^ c, 24)
 	xor0 = v10 ^ v20;
@@ -5476,11 +5536,15 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	v31 = (xor1 >> 16u) ^ (xor0 << 16u);
 
 	// c = c + d
-	o0 = v20 + v30;
-	o1 = v21 + v31;
-	o1 = o1 + select(0u, 1u, o0 < v20);
-	v20 = o0;
-	v21 = o1;
+	v_2021.x = v20;
+	v_2021.y = v21;
+	v_3031.x = v30;
+	v_3031.y = v31;
+	v_2021 = v_2021 + v_3031 + select(vec2<u32>(0u), vec2<u32>(0u, 1u), v_2021.x + v_3031.x < v_2021.x);
+	v20 = v_2021.x;
+	v21 = v_2021.y;
+	v30 = v_3031.x;
+	v31 = v_3031.y;
 
 	// b = rotr64(b ^ c, 63)
 	xor0 = v10 ^ v20;
@@ -6140,11 +6204,15 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	v31 = xor0;
 
 	// c = c + d
-	o0 = v20 + v30;
-	o1 = v21 + v31;
-	o1 = o1 + select(0u, 1u, o0 < v20);
-	v20 = o0;
-	v21 = o1;
+	v_2021.x = v20;
+	v_2021.y = v21;
+	v_3031.x = v30;
+	v_3031.y = v31;
+	v_2021 = v_2021 + v_3031 + select(vec2<u32>(0u), vec2<u32>(0u, 1u), v_2021.x + v_3031.x < v_2021.x);
+	v20 = v_2021.x;
+	v21 = v_2021.y;
+	v30 = v_3031.x;
+	v31 = v_3031.y;
 
 	// b = rotr64(b ^ c, 24)
 	xor0 = v10 ^ v20;
@@ -6177,11 +6245,15 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	v31 = (xor1 >> 16u) ^ (xor0 << 16u);
 
 	// c = c + d
-	o0 = v20 + v30;
-	o1 = v21 + v31;
-	o1 = o1 + select(0u, 1u, o0 < v20);
-	v20 = o0;
-	v21 = o1;
+	v_2021.x = v20;
+	v_2021.y = v21;
+	v_3031.x = v30;
+	v_3031.y = v31;
+	v_2021 = v_2021 + v_3031 + select(vec2<u32>(0u), vec2<u32>(0u, 1u), v_2021.x + v_3031.x < v_2021.x);
+	v20 = v_2021.x;
+	v21 = v_2021.y;
+	v30 = v_3031.x;
+	v31 = v_3031.y;
 
 	// b = rotr64(b ^ c, 63)
 	xor0 = v10 ^ v20;
@@ -6838,11 +6910,15 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	v31 = xor0;
 
 	// c = c + d
-	o0 = v20 + v30;
-	o1 = v21 + v31;
-	o1 = o1 + select(0u, 1u, o0 < v20);
-	v20 = o0;
-	v21 = o1;
+	v_2021.x = v20;
+	v_2021.y = v21;
+	v_3031.x = v30;
+	v_3031.y = v31;
+	v_2021 = v_2021 + v_3031 + select(vec2<u32>(0u), vec2<u32>(0u, 1u), v_2021.x + v_3031.x < v_2021.x);
+	v20 = v_2021.x;
+	v21 = v_2021.y;
+	v30 = v_3031.x;
+	v31 = v_3031.y;
 
 	// b = rotr64(b ^ c, 24)
 	xor0 = v10 ^ v20;
@@ -6878,11 +6954,15 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	v31 = (xor1 >> 16u) ^ (xor0 << 16u);
 
 	// c = c + d
-	o0 = v20 + v30;
-	o1 = v21 + v31;
-	o1 = o1 + select(0u, 1u, o0 < v20);
-	v20 = o0;
-	v21 = o1;
+	v_2021.x = v20;
+	v_2021.y = v21;
+	v_3031.x = v30;
+	v_3031.y = v31;
+	v_2021 = v_2021 + v_3031 + select(vec2<u32>(0u), vec2<u32>(0u, 1u), v_2021.x + v_3031.x < v_2021.x);
+	v20 = v_2021.x;
+	v21 = v_2021.y;
+	v30 = v_3031.x;
+	v31 = v_3031.y;
 
 	// b = rotr64(b ^ c, 63)
 	xor0 = v10 ^ v20;
@@ -7533,11 +7613,15 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	v31 = xor0;
 
 	// c = c + d
-	o0 = v20 + v30;
-	o1 = v21 + v31;
-	o1 = o1 + select(0u, 1u, o0 < v20);
-	v20 = o0;
-	v21 = o1;
+	v_2021.x = v20;
+	v_2021.y = v21;
+	v_3031.x = v30;
+	v_3031.y = v31;
+	v_2021 = v_2021 + v_3031 + select(vec2<u32>(0u), vec2<u32>(0u, 1u), v_2021.x + v_3031.x < v_2021.x);
+	v20 = v_2021.x;
+	v21 = v_2021.y;
+	v30 = v_3031.x;
+	v31 = v_3031.y;
 
 	// b = rotr64(b ^ c, 24)
 	xor0 = v10 ^ v20;
@@ -7573,11 +7657,15 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	v31 = (xor1 >> 16u) ^ (xor0 << 16u);
 
 	// c = c + d
-	o0 = v20 + v30;
-	o1 = v21 + v31;
-	o1 = o1 + select(0u, 1u, o0 < v20);
-	v20 = o0;
-	v21 = o1;
+	v_2021.x = v20;
+	v_2021.y = v21;
+	v_3031.x = v30;
+	v_3031.y = v31;
+	v_2021 = v_2021 + v_3031 + select(vec2<u32>(0u), vec2<u32>(0u, 1u), v_2021.x + v_3031.x < v_2021.x);
+	v20 = v_2021.x;
+	v21 = v_2021.y;
+	v30 = v_3031.x;
+	v31 = v_3031.y;
 
 	// b = rotr64(b ^ c, 63)
 	xor0 = v10 ^ v20;
@@ -8243,11 +8331,15 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	v31 = xor0;
 
 	// c = c + d
-	o0 = v20 + v30;
-	o1 = v21 + v31;
-	o1 = o1 + select(0u, 1u, o0 < v20);
-	v20 = o0;
-	v21 = o1;
+	v_2021.x = v20;
+	v_2021.y = v21;
+	v_3031.x = v30;
+	v_3031.y = v31;
+	v_2021 = v_2021 + v_3031 + select(vec2<u32>(0u), vec2<u32>(0u, 1u), v_2021.x + v_3031.x < v_2021.x);
+	v20 = v_2021.x;
+	v21 = v_2021.y;
+	v30 = v_3031.x;
+	v31 = v_3031.y;
 
 	// b = rotr64(b ^ c, 24)
 	xor0 = v10 ^ v20;
@@ -8283,11 +8375,15 @@ fn main(@builtin(global_invocation_id) id: vec3<u32>) {
 	v31 = (xor1 >> 16u) ^ (xor0 << 16u);
 
 	// c = c + d
-	o0 = v20 + v30;
-	o1 = v21 + v31;
-	o1 = o1 + select(0u, 1u, o0 < v20);
-	v20 = o0;
-	v21 = o1;
+	v_2021.x = v20;
+	v_2021.y = v21;
+	v_3031.x = v30;
+	v_3031.y = v31;
+	v_2021 = v_2021 + v_3031 + select(vec2<u32>(0u), vec2<u32>(0u, 1u), v_2021.x + v_3031.x < v_2021.x);
+	v20 = v_2021.x;
+	v21 = v_2021.y;
+	v30 = v_3031.x;
+	v31 = v_3031.y;
 
 	// b = rotr64(b ^ c, 63)
 	xor0 = v10 ^ v20;
-- 
2.34.1