math/big: add more complete tests and benchmarks of assembly

Also fix a few real but currently harmless bugs from CL 664895.
There were a few places that were still wrong if z != x or if a != 0.

Change-Id: Id8971e2505523bc4708780c82bf998a546f4f081
Reviewed-on: https://go-review.googlesource.com/c/go/+/664897
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Alan Donovan <adonovan@google.com>

Author: rsc

src/math/big/arith_amd64.s

@@ -430,7 +430,7 @@ adx:
 	MOVQ y+48(FP), R8
 	MOVQ m+72(FP), DX
 	MOVQ $0, BX   // i = 0
-	MOVQ $0, CX   // carry
+	MOVQ a+80(FP), CX   // carry
 	CMPQ R11, $8
 	JAE  adx_loop_header
 	CMPQ BX, R11
@@ -446,7 +446,7 @@ adx_loop:
 	MULXQ (R8), SI, DI
 	ADCXQ CX,SI
 	ADOXQ (R10), SI
-	MOVQ  SI,(R10)
+	MOVQ  SI,(R14)
 
 	MULXQ 8(R8), AX, CX
 	ADCXQ DI, AX
@@ -505,7 +505,8 @@ adx_short:
 	MULXQ (R8)(BX*8), SI, DI
 	ADDQ CX, SI
 	ADCQ $0, DI
-	ADDQ SI, (R10)(BX*8)
+	ADDQ (R10)(BX*8), SI
+	MOVQ SI, (R14)(BX*8)
 	ADCQ $0, DI
 	MOVQ DI, CX
 	ADDQ $1, BX		// i++

src/math/big/arith_s390x_test.go

@@ -6,27 +6,10 @@
 
 package big
 
-import (
-	"testing"
-)
+import "testing"
 
-// Tests whether the non vector routines are working, even when the tests are run on a
-// vector-capable machine
-
-func TestFunVVnovec(t *testing.T) {
-	if hasVX {
-		for _, a := range sumVV {
-			arg := a
-			testFunVV(t, "addVV_novec", addVV_novec, arg)
-
-			arg = argVV{a.z, a.y, a.x, a.c}
-			testFunVV(t, "addVV_novec symmetric", addVV_novec, arg)
-
-			arg = argVV{a.x, a.z, a.y, a.c}
-			testFunVV(t, "subVV_novec", subVV_novec, arg)
-
-			arg = argVV{a.y, a.z, a.x, a.c}
-			testFunVV(t, "subVV_novec symmetric", subVV_novec, arg)
-		}
-	}
+func TestNoVec(t *testing.T) {
+	// Make sure non-vector versions match vector versions.
+	t.Run("AddVV", func(t *testing.T) { testVV(t, "addVV_novec", addVV_novec, addVV) })
+	t.Run("SubVV", func(t *testing.T) { testVV(t, "subVV_novec", subVV_novec, subVV) })
 }