Skip a couple unnecessary computations in the final G round. Implement literal of...

diff --git a/src/shaders/compute.wgsl b/src/shaders/compute.wgsl
index 1ee01af3d14a865ae4c9a6116b4525646b2b512f..da0dd968ca8c79ed470df7dc6810bf2003dbb485 100644 (file)
--- a/src/shaders/compute.wgsl
+++ b/src/shaders/compute.wgsl
@@ -20,12 +20,6 @@ struct WORK {
 };
 @group(0) @binding(1) var<storage, read_write> work: WORK;
 
-/**
-* Defined separately from `v0` because the original value is required to
-* calculate the digest and compare it to the threshold.
-*/
-const BLAKE2B_IV32_0: vec2<u32> = vec2(0xF2BDC900u, 0x6A09E667u);
-
 /**
 * Used to rotate bits by a fixed amount during G mixing.
 */
@@ -105,7 +99,7 @@ fn main(id: vec3<u32>) {
        * It is always the "last" compression at this INLEN
        * v14 = ~v14;
        */
-       var v01: vec4<u32> = vec4(BLAKE2B_IV32_0, 0x84CAA73Bu, 0xBB67AE85u);
+       var v01: vec4<u32> = vec4(0xF2BDC900u, 0x6A09E667u, 0x84CAA73Bu, 0xBB67AE85u);
        var v23: vec4<u32> = vec4(0xFE94F82Bu, 0x3C6EF372u, 0x5F1D36F1u, 0xA54FF53Au);
        var v45: vec4<u32> = vec4(0xADE682D1u, 0x510E527Fu, 0x2B3E6C1Fu, 0x9B05688Cu);
        var v67: vec4<u32> = vec4(0xFB41BD6Bu, 0x1F83D9ABu, 0x137E2179u, 0x5BE0CD19u);
@@ -1218,10 +1212,10 @@ fn main(id: vec3<u32>) {
        v01 += vec4(Z, m2) + vec4(Z, 0u, u32(v01.z + m2.x < v01.z));
        v23 += vec4(Z, m3) + vec4(Z, 0u, u32(v23.z + m3.x < v23.z));
 
-       vFC = ((vFC ^ v01) >> ROTATE_16) | ((vFC ^ v01).yxwz << ROTATE_16);
+       // vFC = ((vFC ^ v01) >> ROTATE_16) | ((vFC ^ v01).yxwz << ROTATE_16);
        vDE = ((vDE ^ v23) >> ROTATE_16) | ((vDE ^ v23).yxwz << ROTATE_16);
 
-       vAB += vFC + vec4(0u, u32(vAB.x + vFC.x < vAB.x), 0u, u32(vAB.z + vFC.z < vAB.z));
+       // vAB += vFC + vec4(0u, u32(vAB.x + vFC.x < vAB.x), 0u, u32(vAB.z + vFC.z < vAB.z));
        v89 += vDE + vec4(0u, u32(v89.x + vDE.x < v89.x), 0u, u32(v89.z + vDE.z < v89.z));
 
        // v56 = ((v56 ^ vAB) << ROTATE_1) | ((v56 ^ vAB).yxwz >> ROTATE_31);
@@ -1241,9 +1235,12 @@ fn main(id: vec3<u32>) {
        ****************************************************************************/
 
        /**
-       * Set nonce if it passes the threshold and no other thread has set it
+       * Set nonce if it passes the threshold and no other thread has set it.
+       * Numeric literal used in the finalization digest is the original value of the
+       * first element of the initialization vector `blake2b_IV[0]` which in NanoPow
+       * is initialized at vector component `v01.y`.
        */
-       if ((BLAKE2B_IV32_0.y ^ v01.y ^ v89.y) > ubo.threshold) {
+       if ((0x6A09E667u ^ v01.y ^ v89.y) > ubo.threshold) {
                let wasFound: u32 = atomicExchange(&work.found, 1u);
                if (wasFound == 0u) {
                        work.nonce = m0;