ac/nir: implement nir_op_pack_{us}norm_2x16

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>

diff --git a/src/amd/common/ac_nir_to_llvm.c b/src/amd/common/ac_nir_to_llvm.c
index ec79e5f1ce79ee37499b6e41cdd3d784dcf2fd91..91865bd95ee22dd9c732a6895ae4425ef72c25d5 100644 (file)
--- a/src/amd/common/ac_nir_to_llvm.c
+++ b/src/amd/common/ac_nir_to_llvm.c
@@ -455,8 +455,10 @@ static LLVMValueRef emit_bitfield_select(struct ac_llvm_context *ctx,
                                         LLVMBuildXor(ctx->builder, insert, base, ""), ""), "");
 }
 
-static LLVMValueRef emit_pack_half_2x16(struct ac_llvm_context *ctx,
-                                       LLVMValueRef src0)
+static LLVMValueRef emit_pack_2x16(struct ac_llvm_context *ctx,
+                                  LLVMValueRef src0,
+                                  LLVMValueRef (*pack)(struct ac_llvm_context *ctx,
+                                                       LLVMValueRef args[2]))
 {
        LLVMValueRef comp[2];
 
@@ -464,8 +466,7 @@ static LLVMValueRef emit_pack_half_2x16(struct ac_llvm_context *ctx,
        comp[0] = LLVMBuildExtractElement(ctx->builder, src0, ctx->i32_0, "");
        comp[1] = LLVMBuildExtractElement(ctx->builder, src0, ctx->i32_1, "");
 
-       return LLVMBuildBitCast(ctx->builder, ac_build_cvt_pkrtz_f16(ctx, comp),
-                               ctx->i32, "");
+       return LLVMBuildBitCast(ctx->builder, pack(ctx, comp), ctx->i32, "");
 }
 
 static LLVMValueRef emit_unpack_half_2x16(struct ac_llvm_context *ctx,
@@ -526,6 +527,8 @@ static void visit_alu(struct ac_nir_context *ctx, const nir_alu_instr *instr)
                src_components = 1;
                break;
        case nir_op_pack_half_2x16:
+       case nir_op_pack_snorm_2x16:
+       case nir_op_pack_unorm_2x16:
                src_components = 2;
                break;
        case nir_op_unpack_half_2x16:
@@ -935,7 +938,13 @@ static void visit_alu(struct ac_nir_context *ctx, const nir_alu_instr *instr)
                result = emit_imul_high(&ctx->ac, src[0], src[1]);
                break;
        case nir_op_pack_half_2x16:
-               result = emit_pack_half_2x16(&ctx->ac, src[0]);
+               result = emit_pack_2x16(&ctx->ac, src[0], ac_build_cvt_pkrtz_f16);
+               break;
+       case nir_op_pack_snorm_2x16:
+               result = emit_pack_2x16(&ctx->ac, src[0], ac_build_cvt_pknorm_i16);
+               break;
+       case nir_op_pack_unorm_2x16:
+               result = emit_pack_2x16(&ctx->ac, src[0], ac_build_cvt_pknorm_u16);
                break;
        case nir_op_unpack_half_2x16:
                result = emit_unpack_half_2x16(&ctx->ac, src[0]);

diff --git a/src/gallium/drivers/radeonsi/si_get.c b/src/gallium/drivers/radeonsi/si_get.c
index 6beca2e0744a9bd6dd1d5685aef6fb6d5446b1cb..882bbbbae067bdb9044be78241b7ecc606d90484 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_get.c
+++ b/src/gallium/drivers/radeonsi/si_get.c
@@ -503,9 +503,7 @@ static const struct nir_shader_compiler_options nir_options = {
        .lower_sub = true,
        .lower_ffma = true,
        .lower_fmod = true,
-       .lower_pack_snorm_2x16 = true,
        .lower_pack_snorm_4x8 = true,
-       .lower_pack_unorm_2x16 = true,
        .lower_pack_unorm_4x8 = true,
        .lower_unpack_snorm_2x16 = true,
        .lower_unpack_snorm_4x8 = true,