From d90bb779cae5c5a6754b44680c52cef5e11bd9bf Mon Sep 17 00:00:00 2001
From: Henry de Valence <hdevalence@hdevalence.ca>
Date: Wed, 14 Nov 2018 12:36:34 -0800
Subject: [PATCH] Delay all computations using x[0], y[0]

Since computation of the 0 term in reduction requires a multiplication with a
4-cycle latency, this ensures that the rest of the computation can start before
the 0 term is finished
---
 src/backend/vector/ifma/field.rs | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/src/backend/vector/ifma/field.rs b/src/backend/vector/ifma/field.rs
index 8fdb629..12a72f4 100644
--- a/src/backend/vector/ifma/field.rs
+++ b/src/backend/vector/ifma/field.rs
@@ -197,8 +197,8 @@ impl<'a, 'b> Mul<&'b F51x4Reduced> for &'a F51x4Reduced {
             let mut z9hi = u64x4::splat(0);
 
             // Wave 0
-            z4lo = madd52lo(z4lo, x[4], y[0]);
-            z5hi = madd52hi(z5hi, x[4], y[0]);
+            z4lo = madd52lo(z4lo, x[2], y[2]);
+            z5hi = madd52hi(z5hi, x[2], y[2]);
             z5lo = madd52lo(z5lo, x[4], y[1]);
             z6hi = madd52hi(z6hi, x[4], y[1]);
             z6lo = madd52lo(z6lo, x[4], y[2]);
@@ -219,8 +219,8 @@ impl<'a, 'b> Mul<&'b F51x4Reduced> for &'a F51x4Reduced {
             // Wave 2
             z8lo = madd52lo(z8lo, x[4], y[4]);
             z9hi = madd52hi(z9hi, x[4], y[4]);
-            z4lo = madd52lo(z4lo, x[2], y[2]);
-            z5hi = madd52hi(z5hi, x[2], y[2]);
+            z4lo = madd52lo(z4lo, x[4], y[0]);
+            z5hi = madd52hi(z5hi, x[4], y[0]);
             z5lo = madd52lo(z5lo, x[2], y[3]);
             z6hi = madd52hi(z6hi, x[2], y[3]);
             z6lo = madd52lo(z6lo, x[2], y[4]);