diff options
-rw-r--r-- | src/amd/common/ac_nir_to_llvm.c | 14 | ||||
-rw-r--r-- | src/gallium/drivers/radeonsi/si_shader.c | 8 | ||||
-rw-r--r-- | src/gallium/drivers/radeonsi/si_shader_tgsi_mem.c | 25 | ||||
-rw-r--r-- | src/gallium/drivers/radeonsi/si_shader_tgsi_setup.c | 17 |
4 files changed, 20 insertions, 44 deletions
diff --git a/src/amd/common/ac_nir_to_llvm.c b/src/amd/common/ac_nir_to_llvm.c index 537ac33c044..700e48e14b7 100644 --- a/src/amd/common/ac_nir_to_llvm.c +++ b/src/amd/common/ac_nir_to_llvm.c @@ -486,7 +486,7 @@ static LLVMValueRef emit_unpack_half_2x16(struct ac_llvm_context *ctx, LLVMValueRef src0) { LLVMValueRef const16 = LLVMConstInt(ctx->i32, 16, false); - LLVMValueRef temps[2], result, val; + LLVMValueRef temps[2], val; int i; for (i = 0; i < 2; i++) { @@ -495,12 +495,7 @@ static LLVMValueRef emit_unpack_half_2x16(struct ac_llvm_context *ctx, val = LLVMBuildBitCast(ctx->builder, val, ctx->f16, ""); temps[i] = LLVMBuildFPExt(ctx->builder, val, ctx->f32, ""); } - - result = LLVMBuildInsertElement(ctx->builder, LLVMGetUndef(ctx->v2f32), temps[0], - ctx->i32_0, ""); - result = LLVMBuildInsertElement(ctx->builder, result, temps[1], - ctx->i32_1, ""); - return result; + return ac_build_gather_values(ctx, temps, 2); } static LLVMValueRef emit_ddxy(struct ac_nir_context *ctx, @@ -1004,10 +999,7 @@ static void visit_alu(struct ac_nir_context *ctx, const nir_alu_instr *instr) case nir_op_pack_64_2x32_split: { LLVMValueRef tmp = LLVMGetUndef(ctx->ac.v2i32); - tmp = LLVMBuildInsertElement(ctx->ac.builder, tmp, - src[0], ctx->ac.i32_0, ""); - tmp = LLVMBuildInsertElement(ctx->ac.builder, tmp, - src[1], ctx->ac.i32_1, ""); + tmp = ac_build_gather_values(&ctx->ac, src, 2); result = LLVMBuildBitCast(ctx->ac.builder, tmp, ctx->ac.i64, ""); break; } diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c index 66fe5fad218..cfd99b61601 100644 --- a/src/gallium/drivers/radeonsi/si_shader.c +++ b/src/gallium/drivers/radeonsi/si_shader.c @@ -2724,12 +2724,10 @@ static void emit_streamout_output(struct si_shader_context *ctx, break; case 2: /* as v2i32 */ case 3: /* as v4i32 (aligned to 4) */ + out[3] = LLVMGetUndef(ctx->i32); + /* fall through */ case 4: /* as v4i32 */ - vdata = LLVMGetUndef(LLVMVectorType(ctx->i32, util_next_power_of_two(num_comps))); - for (int j = 0; j < num_comps; j++) { - vdata = LLVMBuildInsertElement(ctx->ac.builder, vdata, out[j], - LLVMConstInt(ctx->i32, j, 0), ""); - } + vdata = ac_build_gather_values(&ctx->ac, out, util_next_power_of_two(num_comps)); break; } diff --git a/src/gallium/drivers/radeonsi/si_shader_tgsi_mem.c b/src/gallium/drivers/radeonsi/si_shader_tgsi_mem.c index 54a0413e464..8e0578b4d5e 100644 --- a/src/gallium/drivers/radeonsi/si_shader_tgsi_mem.c +++ b/src/gallium/drivers/radeonsi/si_shader_tgsi_mem.c @@ -553,7 +553,7 @@ static void store_emit_buffer(struct si_shader_context *ctx, while (writemask) { int start, count; const char *intrinsic_name; - LLVMValueRef data, voff, tmp; + LLVMValueRef data, voff; u_bit_scan_consecutive_range(&writemask, &start, &count); @@ -568,21 +568,14 @@ static void store_emit_buffer(struct si_shader_context *ctx, data = base_data; intrinsic_name = "llvm.amdgcn.buffer.store.v4f32"; } else if (count == 2) { - LLVMTypeRef v2f32 = LLVMVectorType(ctx->f32, 2); - - tmp = LLVMBuildExtractElement( - builder, base_data, - LLVMConstInt(ctx->i32, start, 0), ""); - data = LLVMBuildInsertElement( - builder, LLVMGetUndef(v2f32), tmp, - ctx->i32_0, ""); - - tmp = LLVMBuildExtractElement( - builder, base_data, - LLVMConstInt(ctx->i32, start + 1, 0), ""); - data = LLVMBuildInsertElement( - builder, data, tmp, ctx->i32_1, ""); - + LLVMValueRef values[2] = { + LLVMBuildExtractElement(builder, base_data, + LLVMConstInt(ctx->i32, start, 0), ""), + LLVMBuildExtractElement(builder, base_data, + LLVMConstInt(ctx->i32, start + 1, 0), ""), + }; + + data = ac_build_gather_values(&ctx->ac, values, 2); intrinsic_name = "llvm.amdgcn.buffer.store.v2f32"; } else { assert(count == 1); diff --git a/src/gallium/drivers/radeonsi/si_shader_tgsi_setup.c b/src/gallium/drivers/radeonsi/si_shader_tgsi_setup.c index 1f37b0ba37d..20164939cb7 100644 --- a/src/gallium/drivers/radeonsi/si_shader_tgsi_setup.c +++ b/src/gallium/drivers/radeonsi/si_shader_tgsi_setup.c @@ -305,18 +305,11 @@ si_llvm_emit_fetch_64bit(struct lp_build_tgsi_context *bld_base, LLVMValueRef ptr2) { struct si_shader_context *ctx = si_shader_context(bld_base); - LLVMValueRef result; - - result = LLVMGetUndef(LLVMVectorType(ctx->i32, 2)); - - result = LLVMBuildInsertElement(ctx->ac.builder, - result, - ac_to_integer(&ctx->ac, ptr), - ctx->i32_0, ""); - result = LLVMBuildInsertElement(ctx->ac.builder, - result, - ac_to_integer(&ctx->ac, ptr2), - ctx->i32_1, ""); + LLVMValueRef values[2] = { + ac_to_integer(&ctx->ac, ptr), + ac_to_integer(&ctx->ac, ptr2), + }; + LLVMValueRef result = ac_build_gather_values(&ctx->ac, values, 2); return LLVMBuildBitCast(ctx->ac.builder, result, type, ""); } |