Fix qdmlsl instructions

The qdmlsl instructions were implemented without any saturation.
This has been fixed by utilising existing saturating instructions which
are implemented correctly.

Unit tests have also been updated to test for saturation.

Change-Id: Ia9e5a7bd850bc178920c19e390c17db5a3bfbc4f

arm/neon/qdmlsl.h

@@ -27,13 +27,9 @@
 #if !defined(SIMDE_ARM_NEON_QDMLSL_H)
 #define SIMDE_ARM_NEON_QDMLSL_H
 
-#include "sub.h"
-#include "mul.h"
-#include "mul_n.h"
-#include "movl.h"
-#include "qadd.h"
-#include "qsub.h"
 #include "types.h"
+#include "qsub.h"
+#include "qdmull.h"
 
 HEDLEY_DIAGNOSTIC_PUSH
 SIMDE_DISABLE_UNWANTED_DIAGNOSTICS
@@ -45,7 +41,7 @@ simde_vqdmlslh_s16(int32_t a, int16_t b, int16_t c) {
   #if defined(SIMDE_ARM_NEON_A64V8_NATIVE)
     return vqdmlslh_s16(a, b, c);
   #else
-    return a - HEDLEY_STATIC_CAST(int32_t, b) * HEDLEY_STATIC_CAST(int32_t, c) * 2;
+    return simde_vqsubs_s32(a, simde_vqdmullh_s16(b, c));
   #endif
 }
 #if defined(SIMDE_ARM_NEON_A64V8_ENABLE_NATIVE_ALIASES)
@@ -59,7 +55,7 @@ simde_vqdmlsls_s32(int64_t a, int32_t b, int32_t c) {
   #if defined(SIMDE_ARM_NEON_A64V8_NATIVE)
     return vqdmlsls_s32(a, b, c);
   #else
-    return a - HEDLEY_STATIC_CAST(int64_t, b) * HEDLEY_STATIC_CAST(int64_t, c) * 2;
+    return simde_vqsubd_s64(a, simde_vqdmulls_s32(b, c));
   #endif
 }
 #if defined(SIMDE_ARM_NEON_A64V8_ENABLE_NATIVE_ALIASES)
@@ -73,8 +69,7 @@ simde_vqdmlsl_s16(simde_int32x4_t a, simde_int16x4_t b, simde_int16x4_t c) {
   #if defined(SIMDE_ARM_NEON_A32V7_NATIVE)
     return vqdmlsl_s16(a, b, c);
   #else
-    simde_int32x4_t temp = simde_vmulq_s32(simde_vmovl_s16(b), simde_vmovl_s16(c));
-    return simde_vqsubq_s32(a, simde_vqaddq_s32(temp, temp));
+    return simde_vqsubq_s32(a, simde_vqdmull_s16(b, c));
   #endif
 }
 #if defined(SIMDE_ARM_NEON_A32V7_ENABLE_NATIVE_ALIASES)
@@ -88,10 +83,7 @@ simde_vqdmlsl_s32(simde_int64x2_t a, simde_int32x2_t b, simde_int32x2_t c) {
   #if defined(SIMDE_ARM_NEON_A32V7_NATIVE)
     return vqdmlsl_s32(a, b, c);
   #else
-    simde_int64x2_t r = simde_x_vmulq_s64(
-          simde_vmovl_s32(b),
-          simde_vmovl_s32(c));
-    return simde_vqsubq_s64(a, simde_vqaddq_s64(r, r));
+    return simde_vqsubq_s64(a, simde_vqdmull_s32(b, c));
   #endif
 }
 #if defined(SIMDE_ARM_NEON_A32V7_ENABLE_NATIVE_ALIASES)

arm/neon/qdmlsl_high.h

@@ -28,10 +28,9 @@
 #define SIMDE_ARM_NEON_QDMLSL_HIGH_H
 
 #include "movl_high.h"
-#include "sub.h"
-#include "mul.h"
-#include "mul_n.h"
 #include "types.h"
+#include "qdmull_high.h"
+#include "qsub.h"
 
 HEDLEY_DIAGNOSTIC_PUSH
 SIMDE_DISABLE_UNWANTED_DIAGNOSTICS
@@ -43,7 +42,7 @@ simde_vqdmlsl_high_s16(simde_int32x4_t a, simde_int16x8_t b, simde_int16x8_t c)
   #if defined(SIMDE_ARM_NEON_A64V8_NATIVE)
     return vqdmlsl_high_s16(a, b, c);
   #else
-    return simde_vsubq_s32(a, simde_vmulq_n_s32(simde_vmulq_s32(simde_vmovl_high_s16(b), simde_vmovl_high_s16(c)), 2));
+    return simde_vqsubq_s32(a, simde_vqdmull_high_s16(b, c));
   #endif
 }
 #if defined(SIMDE_ARM_NEON_A64V8_ENABLE_NATIVE_ALIASES)
@@ -57,17 +56,7 @@ simde_vqdmlsl_high_s32(simde_int64x2_t a, simde_int32x4_t b, simde_int32x4_t c)
   #if defined(SIMDE_ARM_NEON_A64V8_NATIVE)
     return vqdmlsl_high_s32(a, b, c);
   #else
-    simde_int64x2_private r_ = simde_int64x2_to_private(
-          simde_x_vmulq_s64(
-          simde_vmovl_high_s32(b),
-          simde_vmovl_high_s32(c)));
-
-    SIMDE_VECTORIZE
-    for (size_t i = 0 ; i < (sizeof(r_.values) / sizeof(r_.values[0])) ; i++) {
-      r_.values[i] = r_.values[i] * HEDLEY_STATIC_CAST(int64_t, 2);
-    }
-
-    return simde_vsubq_s64(a, simde_int64x2_from_private(r_));
+    return simde_vqsubq_s64(a, simde_vqdmull_high_s32(b, c));
   #endif
 }
 #if defined(SIMDE_ARM_NEON_A64V8_ENABLE_NATIVE_ALIASES)

arm/neon/qdmlsl_high_lane.h

@@ -27,91 +27,49 @@
 #if !defined(SIMDE_ARM_NEON_QDMLSL_HIGH_LANE_H)
 #define SIMDE_ARM_NEON_QDMLSL_HIGH_LANE_H
 
-#include "movl_high.h"
-#include "sub.h"
-#include "mul.h"
-#include "mul_n.h"
-#include "dup_n.h"
+#include "dup_lane.h"
+#include "get_high.h"
 #include "types.h"
+#include "qdmlsl.h"
 
 HEDLEY_DIAGNOSTIC_PUSH
 SIMDE_DISABLE_UNWANTED_DIAGNOSTICS
 SIMDE_BEGIN_DECLS_
 
-SIMDE_FUNCTION_ATTRIBUTES
-simde_int32x4_t
-simde_vqdmlsl_high_lane_s16(simde_int32x4_t a, simde_int16x8_t b, simde_int16x4_t v, const int lane) SIMDE_REQUIRE_CONSTANT_RANGE(lane, 0, 3) {
-    return simde_vsubq_s32(a,
-        simde_vmulq_n_s32(
-        simde_vmulq_s32(
-        simde_vmovl_high_s16(b),
-        simde_vmovl_high_s16(simde_vdupq_n_s16(simde_int16x4_to_private(v).values[lane]))), 2));
-}
 #if defined(SIMDE_ARM_NEON_A64V8_NATIVE)
   #define simde_vqdmlsl_high_lane_s16(a, b, v, lane) vqdmlsl_high_lane_s16(a, b, v, lane)
+#else
+  #define simde_vqdmlsl_high_lane_s16(a, b, v, lane) simde_vqdmlsl_s16((a), simde_vget_high_s16((b)), simde_vdup_lane_s16((v), (lane)))
 #endif
 #if defined(SIMDE_ARM_NEON_A64V8_ENABLE_NATIVE_ALIASES)
   #undef vqdmlsl_high_lane_s16
   #define vqdmlsl_high_lane_s16(a, b, v, lane) simde_vqdmlsl_high_lane_s16((a), (b), (v), (lane))
 #endif
 
-SIMDE_FUNCTION_ATTRIBUTES
-simde_int32x4_t
-simde_vqdmlsl_high_laneq_s16(simde_int32x4_t a, simde_int16x8_t b, simde_int16x8_t v, const int lane) SIMDE_REQUIRE_CONSTANT_RANGE(lane, 0, 7) {
-    return simde_vsubq_s32(a,
-        simde_vmulq_n_s32(
-        simde_vmulq_s32(
-        simde_vmovl_high_s16(b),
-        simde_vmovl_high_s16(simde_vdupq_n_s16(simde_int16x8_to_private(v).values[lane]))), 2));
-}
 #if defined(SIMDE_ARM_NEON_A64V8_NATIVE)
   #define simde_vqdmlsl_high_laneq_s16(a, b, v, lane) vqdmlsl_high_laneq_s16(a, b, v, lane)
+#else
+  #define simde_vqdmlsl_high_laneq_s16(a, b, v, lane) simde_vqdmlsl_s16((a), simde_vget_high_s16((b)), simde_vdup_laneq_s16((v), (lane)))
 #endif
 #if defined(SIMDE_ARM_NEON_A64V8_ENABLE_NATIVE_ALIASES)
   #undef vqdmlsl_high_laneq_s16
   #define vqdmlsl_high_laneq_s16(a, b, v, lane) simde_vqdmlsl_high_laneq_s16((a), (b), (v), (lane))
 #endif
 
-SIMDE_FUNCTION_ATTRIBUTES
-simde_int64x2_t
-simde_vqdmlsl_high_lane_s32(simde_int64x2_t a, simde_int32x4_t b, simde_int32x2_t v, const int lane) SIMDE_REQUIRE_CONSTANT_RANGE(lane, 0, 1) {
-  simde_int64x2_private r_ = simde_int64x2_to_private(
-        simde_x_vmulq_s64(
-        simde_vmovl_high_s32(b),
-        simde_vmovl_high_s32(simde_vdupq_n_s32(simde_int32x2_to_private(v).values[lane]))));
-
-  SIMDE_VECTORIZE
-  for (size_t i = 0 ; i < (sizeof(r_.values) / sizeof(r_.values[0])) ; i++) {
-    r_.values[i] = r_.values[i] * HEDLEY_STATIC_CAST(int64_t, 2);
-  }
-
-  return simde_vsubq_s64(a, simde_int64x2_from_private(r_));
-}
 #if defined(SIMDE_ARM_NEON_A64V8_NATIVE)
   #define simde_vqdmlsl_high_lane_s32(a, b, v, lane) vqdmlsl_high_lane_s32(a, b, v, lane)
+#else
+  #define simde_vqdmlsl_high_lane_s32(a, b, v, lane) simde_vqdmlsl_s32((a), simde_vget_high_s32((b)), simde_vdup_lane_s32((v), (lane)))
 #endif
 #if defined(SIMDE_ARM_NEON_A64V8_ENABLE_NATIVE_ALIASES)
   #undef vqdmlsl_high_lane_s32
   #define vqdmlsl_high_lane_s32(a, b, v, lane) simde_vqdmlsl_high_lane_s32((a), (b), (v), (lane))
 #endif
 
-SIMDE_FUNCTION_ATTRIBUTES
-simde_int64x2_t
-simde_vqdmlsl_high_laneq_s32(simde_int64x2_t a, simde_int32x4_t b, simde_int32x4_t v, const int lane) SIMDE_REQUIRE_CONSTANT_RANGE(lane, 0, 3) {
-  simde_int64x2_private r_ = simde_int64x2_to_private(
-        simde_x_vmulq_s64(
-        simde_vmovl_high_s32(b),
-        simde_vmovl_high_s32(simde_vdupq_n_s32(simde_int32x4_to_private(v).values[lane]))));
-
-  SIMDE_VECTORIZE
-  for (size_t i = 0 ; i < (sizeof(r_.values) / sizeof(r_.values[0])) ; i++) {
-    r_.values[i] = r_.values[i] * HEDLEY_STATIC_CAST(int64_t, 2);
-  }
-
-  return simde_vsubq_s64(a, simde_int64x2_from_private(r_));
-}
 #if defined(SIMDE_ARM_NEON_A64V8_NATIVE)
   #define simde_vqdmlsl_high_laneq_s32(a, b, v, lane) vqdmlsl_high_laneq_s32(a, b, v, lane)
+#else
+  #define simde_vqdmlsl_high_laneq_s32(a, b, v, lane) simde_vqdmlsl_s32((a), simde_vget_high_s32((b)), simde_vdup_laneq_s32((v), (lane)))
 #endif
 #if defined(SIMDE_ARM_NEON_A64V8_ENABLE_NATIVE_ALIASES)
   #undef vqdmlsl_high_laneq_s32

arm/neon/qdmlsl_high_n.h

@@ -29,10 +29,8 @@
 
 #include "movl_high.h"
 #include "dup_n.h"
-#include "sub.h"
-#include "mul.h"
-#include "mul_n.h"
 #include "types.h"
+#include "qdmlsl_high.h"
 
 HEDLEY_DIAGNOSTIC_PUSH
 SIMDE_DISABLE_UNWANTED_DIAGNOSTICS
@@ -44,11 +42,7 @@ simde_vqdmlsl_high_n_s16(simde_int32x4_t a, simde_int16x8_t b, int16_t c) {
   #if defined(SIMDE_ARM_NEON_A64V8_NATIVE)
     return vqdmlsl_high_n_s16(a, b, c);
   #else
-    return simde_vsubq_s32(a,
-        simde_vmulq_n_s32(
-        simde_vmulq_s32(
-        simde_vmovl_high_s16(b),
-        simde_vmovl_high_s16(simde_vdupq_n_s16(c))), 2));
+    return simde_vqdmlsl_high_s16(a, b, simde_vdupq_n_s16(c));
   #endif
 }
 #if defined(SIMDE_ARM_NEON_A64V8_ENABLE_NATIVE_ALIASES)
@@ -62,17 +56,7 @@ simde_vqdmlsl_high_n_s32(simde_int64x2_t a, simde_int32x4_t b, int32_t c) {
   #if defined(SIMDE_ARM_NEON_A64V8_NATIVE)
     return vqdmlsl_high_n_s32(a, b, c);
   #else
-    simde_int64x2_private r_ = simde_int64x2_to_private(
-          simde_x_vmulq_s64(
-          simde_vmovl_high_s32(b),
-          simde_vmovl_high_s32(simde_vdupq_n_s32(c))));
-
-    SIMDE_VECTORIZE
-    for (size_t i = 0 ; i < (sizeof(r_.values) / sizeof(r_.values[0])) ; i++) {
-      r_.values[i] = r_.values[i] * HEDLEY_STATIC_CAST(int64_t, 2);
-    }
-
-    return simde_vsubq_s64(a, simde_int64x2_from_private(r_));
+    return simde_vqdmlsl_high_s32(a, b, simde_vdupq_n_s32(c));
   #endif
 }
 #if defined(SIMDE_ARM_NEON_A64V8_ENABLE_NATIVE_ALIASES)