Auto merge of #32140 - ruud-v-a:avx-intrinsics, r=alexcrichton

Add AVX broadcast and conversion intrinsics

This adds the following intrinsics:

 * `_mm256_broadcast_pd`
 * `_mm256_broadcast_ps`
 * `_mm256_cvtepi32_pd`
 * `_mm256_cvtepi32_ps`
 * `_mm256_cvtpd_epi32`
 * `_mm256_cvtpd_ps`
 * `_mm256_cvtps_epi32`
 * `_mm256_cvtps_pd`
 * `_mm256_cvttpd_epi32`
 * `_mm256_cvttps_epi32`

The "avx" codegen feature must be enabled to use these.

2 files changed, 113 insertions, 0 deletions

diff --git a/src/etc/platform-intrinsics/x86/avx.json b/src/etc/platform-intrinsics/x86/avx.json
index 2c1492c2954..08524fbd6dd 100644
--- a/src/etc/platform-intrinsics/x86/avx.json
+++ b/src/etc/platform-intrinsics/x86/avx.json
@@ -9,6 +9,69 @@
             "args": ["0", "0"]
         },
         {
+            "intrinsic": "256_broadcast_{0.data_type}",
+            "width": [256],
+            "llvm": "vbroadcastf128.{0.data_type}.256",
+            "ret": "f(32-64)",
+            "args": ["s8SPc"]
+        },
+        {
+            "intrinsic": "256_cvtepi32_pd",
+            "width": [256],
+            "llvm": "cvtdq2.pd.256",
+            "ret": "f64",
+            "args": ["s32h"]
+        },
+        {
+            "intrinsic": "256_cvtepi32_ps",
+            "width": [256],
+            "llvm": "cvtdq2.ps.256",
+            "ret": "f32",
+            "args": ["s32"]
+        },
+        {
+            "intrinsic": "256_cvtpd_epi32",
+            "width": [256],
+            "llvm": "cvt.pd2dq.256",
+            "ret": "s32h",
+            "args": ["f64"]
+        },
+        {
+            "intrinsic": "256_cvtpd_ps",
+            "width": [256],
+            "llvm": "cvt.pd2.ps.256",
+            "ret": "f32h",
+            "args": ["f64"]
+        },
+        {
+            "intrinsic": "256_cvtps_epi32",
+            "width": [256],
+            "llvm": "cvt.ps2dq.256",
+            "ret": "s32",
+            "args": ["f32"]
+        },
+        {
+            "intrinsic": "256_cvtps_pd",
+            "width": [256],
+            "llvm": "cvt.ps2.pd.256",
+            "ret": "f64",
+            "args": ["f32h"]
+        },
+        {
+            "intrinsic": "256_cvttpd_epi32",
+            "width": [256],
+            "llvm": "cvtt.pd2dq.256",
+            "ret": "s32h",
+            "args": ["f64"]
+        },
+        {
+            "intrinsic": "256_cvttps_epi32",
+            "width": [256],
+            "llvm": "cvtt.ps2dq.256",
+            "ret": "s32",
+            "args": ["f32"]
+        },
+        {
             "intrinsic": "256_dp_ps",
             "width": [256],
             "llvm": "dp.ps.256",
diff --git a/src/librustc_platform_intrinsics/x86.rs b/src/librustc_platform_intrinsics/x86.rs
index 168ae79ab74..d8aaf151267 100644
--- a/src/librustc_platform_intrinsics/x86.rs
+++ b/src/librustc_platform_intrinsics/x86.rs
@@ -498,6 +498,56 @@ pub fn find<'tcx>(_tcx: &TyCtxt<'tcx>, name: &str) -> Option<Intrinsic> {
             output: v(f(64), 4),
             definition: Named("llvm.x86.avx.addsub.pd.256")
         },
+        "256_broadcast_ps" => Intrinsic {
+            inputs: vec![p(true, i(8), None)],
+            output: v(f(32), 8),
+            definition: Named("llvm.x86.avx.vbroadcastf128.ps.256")
+        },
+        "256_broadcast_pd" => Intrinsic {
+            inputs: vec![p(true, i(8), None)],
+            output: v(f(64), 4),
+            definition: Named("llvm.x86.avx.vbroadcastf128.pd.256")
+        },
+        "256_cvtepi32_pd" => Intrinsic {
+            inputs: vec![v(i(32), 4)],
+            output: v(f(64), 4),
+            definition: Named("llvm.x86.avx.cvtdq2.pd.256")
+        },
+        "256_cvtepi32_ps" => Intrinsic {
+            inputs: vec![v(i(32), 8)],
+            output: v(f(32), 8),
+            definition: Named("llvm.x86.avx.cvtdq2.ps.256")
+        },
+        "256_cvtpd_epi32" => Intrinsic {
+            inputs: vec![v(f(64), 4)],
+            output: v(i(32), 4),
+            definition: Named("llvm.x86.avx.cvt.pd2dq.256")
+        },
+        "256_cvtpd_ps" => Intrinsic {
+            inputs: vec![v(f(64), 4)],
+            output: v(f(32), 4),
+            definition: Named("llvm.x86.avx.cvt.pd2.ps.256")
+        },
+        "256_cvtps_epi32" => Intrinsic {
+            inputs: vec![v(f(32), 8)],
+            output: v(i(32), 8),
+            definition: Named("llvm.x86.avx.cvt.ps2dq.256")
+        },
+        "256_cvtps_pd" => Intrinsic {
+            inputs: vec![v(f(32), 4)],
+            output: v(f(64), 4),
+            definition: Named("llvm.x86.avx.cvt.ps2.pd.256")
+        },
+        "256_cvttpd_epi32" => Intrinsic {
+            inputs: vec![v(f(64), 4)],
+            output: v(i(32), 4),
+            definition: Named("llvm.x86.avx.cvtt.pd2dq.256")
+        },
+        "256_cvttps_epi32" => Intrinsic {
+            inputs: vec![v(f(32), 8)],
+            output: v(i(32), 8),
+            definition: Named("llvm.x86.avx.cvtt.ps2dq.256")
+        },
         "256_dp_ps" => Intrinsic {
             inputs: vec![v(f(32), 8), v(f(32), 8), i_(32, 8)],
             output: v(f(32), 8),