commit e1e66707196545b787c3002fb86ca1b33b9f962a
author George Steed <george.steed@arm.com> Fri Nov 03 07:12:02 2023 +0000
committer James Zern <jzern@google.com> Fri Nov 03 17:29:16 2023 +0000
tree b2830c637624c9b449f49dc8d97daa8e3ab04f33
parent d0f3147e5242edf52b896a65b23af47006e9e84f

fwd_txfm_neon.c: Avoid tran_high_t->int32_t narrowing warnings

The cospi constants are defined as tran_high_t which is typedef'd as
int64_t. None of the constants exceed the range of int32_t though so
casting them here is fine.

Bug: aomedia:3507
Change-Id: I84df4353da72a98e08f4fbcef9f1339dd54098de

aom_dsp/arm/fwd_txfm_neon.c

diff --git a/aom_dsp/arm/fwd_txfm_neon.c b/aom_dsp/arm/fwd_txfm_neon.c
index 4746c28..fb4cda7 100644
--- a/aom_dsp/arm/fwd_txfm_neon.c
+++ b/aom_dsp/arm/fwd_txfm_neon.c
@@ -48,8 +48,8 @@
     // Must expand all elements to s32. See 'needs32' comment in fwd_txfm.c.
     const int32x4_t s_0_p_s_1 = vaddl_s16(s_0, s_1);
     const int32x4_t s_0_m_s_1 = vsubl_s16(s_0, s_1);
-    const int32x4_t temp1 = vmulq_n_s32(s_0_p_s_1, cospi_16_64);
-    const int32x4_t temp2 = vmulq_n_s32(s_0_m_s_1, cospi_16_64);
+    const int32x4_t temp1 = vmulq_n_s32(s_0_p_s_1, (int32_t)cospi_16_64);
+    const int32x4_t temp2 = vmulq_n_s32(s_0_m_s_1, (int32_t)cospi_16_64);
 
     // fdct_round_shift
     int16x4_t out_0 = vrshrn_n_s32(temp1, DCT_CONST_BITS);
@@ -57,11 +57,13 @@
 
     // s_3 * cospi_8_64 + s_2 * cospi_24_64
     // s_3 * cospi_24_64 - s_2 * cospi_8_64
-    const int32x4_t s_3_cospi_8_64 = vmull_n_s16(s_3, cospi_8_64);
-    const int32x4_t s_3_cospi_24_64 = vmull_n_s16(s_3, cospi_24_64);
+    const int32x4_t s_3_cospi_8_64 = vmull_n_s16(s_3, (int32_t)cospi_8_64);
+    const int32x4_t s_3_cospi_24_64 = vmull_n_s16(s_3, (int32_t)cospi_24_64);
 
-    const int32x4_t temp3 = vmlal_n_s16(s_3_cospi_8_64, s_2, cospi_24_64);
-    const int32x4_t temp4 = vmlsl_n_s16(s_3_cospi_24_64, s_2, cospi_8_64);
+    const int32x4_t temp3 =
+        vmlal_n_s16(s_3_cospi_8_64, s_2, (int32_t)cospi_24_64);
+    const int32x4_t temp4 =
+        vmlsl_n_s16(s_3_cospi_24_64, s_2, (int32_t)cospi_8_64);
 
     // fdct_round_shift
     int16x4_t out_1 = vrshrn_n_s32(temp3, DCT_CONST_BITS);