radv: use a faster version for nir_op_pack_half_2x16

This patch is ported from RadeonSI and it has two effects.

It fixes a rendering issue which affects F1 2017 and Dawn
of War 3 (Vega only) because LLVM was ending up by generating
the new v_mad_mix_{hi,lo} instructions which appear to be
buggy in some way. Not sure if Mesa is generating something
wrong or if the issue is in LLVM only. Anyway, that explains why
the DOW3 issue can't be reproduced with GL on Vega.

It also improves performance because v_cvt_pkrtz_f16 is faster,
and because I guess the rounding mode behaviour is similar between
GL and VK, we can use it. About performance, it improves Talos
by +3/4% but I don't see any other impacts.

No CTS regressions on Polaris.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>

diff --git a/src/amd/common/ac_nir_to_llvm.c b/src/amd/common/ac_nir_to_llvm.c
index 96ba289a813589ed6adfbd60246f84465992c8a8..663b27d265a5b3a71843e147855b51353d831a3e 100644 (file)
--- a/src/amd/common/ac_nir_to_llvm.c
+++ b/src/amd/common/ac_nir_to_llvm.c
@@ -1426,23 +1426,13 @@ static LLVMValueRef emit_bitfield_insert(struct ac_llvm_context *ctx,
 static LLVMValueRef emit_pack_half_2x16(struct ac_llvm_context *ctx,
                                        LLVMValueRef src0)
 {
-       LLVMValueRef const16 = LLVMConstInt(ctx->i32, 16, false);
-       int i;
        LLVMValueRef comp[2];
 
        src0 = ac_to_float(ctx, src0);
        comp[0] = LLVMBuildExtractElement(ctx->builder, src0, ctx->i32_0, "");
        comp[1] = LLVMBuildExtractElement(ctx->builder, src0, ctx->i32_1, "");
-       for (i = 0; i < 2; i++) {
-               comp[i] = LLVMBuildFPTrunc(ctx->builder, comp[i], ctx->f16, "");
-               comp[i] = LLVMBuildBitCast(ctx->builder, comp[i], ctx->i16, "");
-               comp[i] = LLVMBuildZExt(ctx->builder, comp[i], ctx->i32, "");
-       }
-
-       comp[1] = LLVMBuildShl(ctx->builder, comp[1], const16, "");
-       comp[0] = LLVMBuildOr(ctx->builder, comp[0], comp[1], "");
 
-       return comp[0];
+       return ac_build_cvt_pkrtz_f16(ctx, comp);
 }
 
 static LLVMValueRef emit_unpack_half_2x16(struct ac_llvm_context *ctx,